Raman Spectroscopy

in Cancer Diagnostics and Surgery: 25 Years of Progress from Surface-Enhanced Raman Spectroscopy to Artificial IntelligenceA Bibliometric and Visualized Study

Authors: Chong Ma, Haitao Zhou, Jingfei Teng, Ruonan Wang, Fuguang Zhao, Wenjin Zhou, Yawei Guan, Zhifang Wu, Sijin Li, Xing Ai, Liantuan Xiao

Abstract

Introduction

Cancer remains one of the most formidable challenges in modern medicine, with its early detection and accurate diagnosis being critical to improving patient outcomes. , Traditional diagnostic methods, such as imaging techniques (e.g., ultrasound, computed tomography (CT), and Magnetic Resonance Imaging (MRI)) and biomarker assays, often suffer from limitations in sensitivity and specificity. Histopathology, considered the gold standard, is inherently invasive, time-consuming, and prone to sampling errors. Moreover, the heterogeneity of cancer cells can lead to missed diagnoses, incomplete tumor characterization, and suboptimal treatment planning. Beyond early diagnosis, a particularly critical challenge in cancer treatment is the accurate determination of tumor margins during surgery. Achieving complete resection of the tumor while preserving healthy tissue is essential for reducing the risk of recurrence and minimizing postoperative complications. Current methods for intraoperative margin assessment, such as frozen section analysis, are time-consuming and can be unreliable, particularly in complex or heterogeneous tumors. This challenge is especially pronounced in surgeries for cancers of the brain, breast, and prostate, where preserving functional tissue is crucial for the patient’s quality of life. , These challenges highlight the urgent need for rapid, noninvasive, and highly accurate technologies that can overcome the limitations of existing approaches and revolutionize cancer detection.

Raman spectroscopy, an optical technique based on the inelastic scattering of light, has emerged as a potential tool for addressing many of the challenges in cancer detection. Although the inelastic scattering of light had been theoretically predicted by Smekal in 1923, the first experimental demonstrations were reported in 1928 by C.V. Raman and K.S. Krishnan working in India and, almost simultaneously, by G.S. Landsberg and L.I. Mandelstam. Raman and Krishnan observed this phenomenon in liquids, reporting that a fraction of scattered monochromatic light undergoes frequency shifts due to molecular vibrationsa phenomenon now termed Raman scattering. The Nobel Prize in Physics in 1930 was awarded to C.V. Raman for this discovery. This groundbreaking work earned the 1930 Nobel Prize in Physics and established a noninvasive molecular analysis paradigm. Unlike infrared spectroscopy relying on dipole moment changes, Raman effects originate from molecular polarizability variations, enabling effective detection of nonpolar molecules and aqueous samples. Advantages including minimal sample preparation and compatibility with encapsulated materials propelled Raman technology from laboratories to industrial inspection and clinical diagnostics. , By measuring the vibrational modes of molecules, Raman spectroscopy generates a unique spectral fingerprint that reflects the biochemical makeup of the sample. This capability enables the detection of subtle molecular changes associated with cancer, often at a very early stage. Like infrared (IR) absorption spectroscopy, Raman spectroscopy probes molecular vibrational modes and generates structural fingerprint information. However, their selection rules IR absorption requires a change in the electric dipole moment, whereas Raman scattering requires a change in molecular polarizability. As a result, the two techniques are often complementary. A key practical advantage of Raman spectroscopy for medical applications is its relative insensitivity to water, which strongly absorbs IR radiation and thus complicates IR-based measurements of biological samples.

A key advantage of Raman spectroscopy is the noninvasive nature. The technique can be applied to intact tissues and in vivo. This feature makes it particularly suitable for real-time applications, such as intraoperative tumor margin assessment and endoscopic diagnostics. Additionally, the sensitivity of Raman spectroscopy can be further enhanced by advanced variants of the technique, including surface-enhanced Raman spectroscopy (SERS), coherent anti-Stokes Raman scattering (CARS), , and stimulated Raman scattering (SRS), each developed at different stages of the field’s history (see Discussion Section 2 for a detailed historical overview). For example, SERS nanoparticles functionalized with targeting ligands can selectively bind to cancer cells, enabling highly sensitive and specific detection of tumors in vivo. Raman spectroscopy has also shown great promise in liquid biopsy applications. By analyzing biofluids such as blood, urine, and saliva, the technique can detect circulating tumor cells (CTCs), exosomes, and other cancer-derived biomarkers. , The integration of Raman spectroscopy with advanced computational methods and artificial intelligence (AI) has further expanded its potential. These tools enable the rapid analysis of complex spectral data, improving diagnostic accuracy and facilitating the discovery of novel biomarkers. Raman spectroscopy is also increasingly used to guide cancer treatment. Raman-guided photothermal and photodynamic therapies, which use light-activated nanoparticles to selectively destroy cancer cells, represent another exciting frontier. These therapies leverage the precision of Raman spectroscopy to target tumors with minimal off-target effects, offering a promising alternative to conventional treatments.

The aim of this study is to provide a quantitative and comprehensive analysis of the advancements in Raman spectroscopy over the past quarter century, with a particular focus on its applications in oncology. The year 2024 marks the 50th anniversary of SERS discovery. By synthesizing the vast body of literature published since 2000, we seek to map the evolution of the field, identify key trends and breakthroughs, and highlight the challenges and opportunities that lie ahead. By combining quantitative data with visualization, we aim to provide a holistic perspective on the past, present, and future of Raman spectroscopy in oncology.

Methods

Data Source and Retrieval Strategy

This bibliometric study utilized data from the Web of Science Core Collection (WOSCC), which is a comprehensive and authoritative database for scientific publications. The retrieval strategy was designed to capture as many relevant studies as possible on the application of Raman optical technologies in tumor research. Synonyms for tumor, singular, and plural forms were considered and included. The retrieval query was defined as TS = (Raman) AND ((TS = (cancer*) OR TS = (tumor*) OR TS = (carcinoma*) OR TS = (neoplasm*)).

Inclusion criteria: (1) publication time January 2000 to December 2024; (2) document “article” or “review;” (3) English. Exclusion (1) duplicate; (2) retracted or publication with expression of concern. This strategy ensured the inclusion of high-quality publications while excluding other document types such as conference proceedings, editorials, and letters. The search was conducted on fifth February 2025, and the data were exported in plain text format with full records and cited references.

Data Extraction

The retrieved data were imported into CiteSpace (version 6.4.R1, 64 bit, advanced). The data set included publication titles, authors, affiliations, keywords, abstracts, citation counts, and reference lists. When conducting bibliometric analysis using CiteSpace, the terms “keyword” and “term” have different definitions and usages. “Keyword” is based on the keywords provided directly by the authors of the literature. The results align more closely with the authors’ intentions, making them suitable for quickly understanding the core themes of a field. “Term” is extracted from the titles and abstracts of the literature, identifying high-frequency words or phrases through algorithms, including terms not present in the keywords. “Term” analysis is more comprehensive, potentially revealing implicit research connections and directions. We conducted both “keywords” and “term” analysis. To merge synonymous terms such as “surface-enhanced Raman spectroscopy,” “surface enhanced Raman spectroscopy,” and “SERS,” a preprocessing step was applied to the WOS data. In cases of discrepancies in data extraction from the included studies, two researchers resolved differences through discussion.

Analysis and Visualization

Study

Subjects

Publication and citation trends were plotted annually to identify growth patterns in the field. Countries/Regions’ outputs, collaboration networks, and geographic distributions were visualized using network map overlaying world map. Institutions were analyzed based on publication counts and citations. Authors and cited authors were analyzed to identify most productive and influential researchers and their cooperation. References with citation burst were detected to identify seminal and milestone works. Keywords and terms analysis and clustering were performed to identify research hotspots and emerging trends. Categories and dual-overlay mapping were analyzed to understand interdisciplinary contributions.

Parameters and Visualization

In the network visualizations produced by CiteSpace, nodes represented specific parameters (e.g., countries, institutions, keywords), with node size indicating frequency. The tree ring color of each node is a visual feature used to represent temporal information about the node, with outer rings representing more recent time periods. And connections between nodes indicated collaboration or co-occurrence. H-Index is the largest number h such that h publications of an author have at least h citations each. Higher H-index indicates greater research impact and productivity. Betweenness centrality (BC) was used to identify key nodes that acted as bridges in the network, with a BC value greater than 0.1 signifying high influence. The modularity score (Q score) and silhouette score (S score) were used to assess the structure and reliability of keyword clusters. A Q score greater than 0.3 indicated a well-structured network, while an S score greater than 0.7 suggested high reliability. Burst detection was employed to identify a sharp increase in frequency over a short period, revealing research hotspots and frontiers. Timeline visualization was used to map the evolution of research themes over time. Bradford’s Law is a bibliometric principle that describes the distribution of scientific literature across journals.

Results

Scientific Output

There were 9994 results related to the Raman technique and tumor research from the Web of Science Core Collection (WOSCC) on 2025 February 5. After screening with the criteria in the Methods section above, a total of 8667 publications (7526 articles and 1141 reviews) were included for further analysis, as shown in Figure S1. The total of citations for these publications was 323,507, with an average of 37.3 citations per publication, indicating that this field has attracted significant attention. The annual publication output exhibited a steady increase, with an annual growth rate of 20.73%, reflecting growing interest in Raman spectroscopy for cancer research (Figure ). The cumulative publication counts surpassed 100 publications in 2004, 500 in 2009, 1000 in 2012, and 5000 in 2020. The cumulative publications followed a second-degree polynomial distribution (y = 22.277x^2^ – 239.99x + 606.74, R ^2^ = 0.996), indicating rapid growth in the field (Figure S2). For citations, the growth rate was even more rapid as a power function (y = 1.6675x^3.7674^, R ^2^ = 0.999), indicating the extensive influence and appeal of this field (Figure S3).

Countries/Regions

A total of 108 countries and regions contributed to the global scientific output in this field. As shown in Figure , publications in Raman technology for cancer research were primarily geographically distributed across East and South Asia, North America, and Europe. The top 10 countries/regions by publication count are presented in Table S1. China ranked first with 2550 publications, followed by the USA (1629 publications) and India (1032 publications). It is not surprising that countries with larger populations tend to publish more relevant papers. To better reflect relative research activity, we calculated the number of papers per capita for countries and regions that have published more than 100 papers in this field. Our results showed that Ireland (29.47/million people), Singapore (27.19/million people), and Scotland (21.09/million people) ranked among the top.

Nevertheless, the United States has the highest citation counts. In terms of the H-index, the USA led with 148, followed by China (118) and England (80). The betweenness centrality (BC) analysis revealed that India (BC = 0.32), the USA (BC = 0.15), and France (BC = 0.12) were the most central nodes in the collaboration network, indicating their roles in international cooperation.

Institutions

A total of 5849 institutions contributed to the field, with the Chinese Academy of Sciences leading with 323 publications, followed by Centre National de la Recherche Scientifique (CNRS) in France (181 publications) and Fujian Normal University in China (176 publications). In terms of the H-index, the top institutions were the Chinese Academy of Sciences (63), University of California System (48), and National University of Singapore (45). The collaboration network highlighted the Chinese Academy of Sciences (BC = 0.25), Centre National de la Recherche Scientifique (BC = 0.18), and University of California System (BC = 0.16) as the most central institutions. The top 10 institutions by publication count are presented in Table S2. The timeline map highlights the temporal evolution of research output from key institutions and their focus areas (Figure ). The cluster of “surface-enhanced Raman spectroscopy” has remained a focal point of sustained interest, attracting numerous high-output institutions like the Chinese Academy of Sciences. Centre National de la Recherche Scientifique was one of the top institutions to apply Raman techniques in biological tissues. The University of California System has made significant contributions to the development of SERS nanoparticles. More recently, the Polish Academy of Sciences has investigated the application of Raman techniques in prostate cancer.

Author and Cocited Author

At the end of 2024, there were 29,519 authors that had contributed to the publication within the field of Raman spectroscopy and tumor research. Through the analysis of publication counts and collaboration networks, we identified 23 active authors with more than 50 publications. The top 10 authors in terms of counts are as listed in Table , with Nicholas Stone (92 publications, H-index = 39), Shangyuan Feng (90 publications, H-index = 33), and Jürgen Popp (82 publications, H-index = 33) ranking first, second, and third, respectively. Burst detection analysis identified authors who have made significant impacts at different stages of the field (Figure S4). Zhiwei Huang, Haishan Zeng, and Wei Zheng emerged early, indicating their pioneering contributions during the formative years of Raman spectroscopy in tumor research. Then Rong Chen, Hugh Byrne, and Jürgen Popp have exerted a sustained and bridging role during key developmental stages of this field. More recently, Xiaowei Cao and Jian Ye integrated artificial intelligence with advanced Raman optics for cancer detection, achieving a strong burst intensity. The timeline cluster analysis of authors (small clusters were filtering out) indicates a very robust structured and reliable network (modularity Q = 0.9754; silhouette S = 0.99). As shown in Figure , the largest cluster was “surface-enhanced Raman spectroscopy,” including a number of top authors in this field like Shangyuan Feng and Haishan Zeng. “Optical pathology” represents one of the most significant research directions in integrating Raman spectroscopy with clinical oncology, aimed at enhancing both the accuracy and timeliness of traditional histopathological diagnosisthe current gold standard in tumor diagnosis. Murali Krishna and Arti Hole are seminal contributors to this field. Another advanced variant of conventional Raman spectroscopy“resonance Raman”has also been a sustained attractive subdirection. Notably, in CiteSpace clustering, an author is assigned to only one cluster corresponding to their most distinctive research direction, even if they have contributed to multiple research directions. When interpreting the clustering map, it should be emphasized that highly productive authors such as Nicholas Stone, Jürgen Popp, and Hugh Byrne have made significant contributions across directions.

Cocited authors are authors who are frequently cited together in the reference lists of other papers. The CiteSpace burst analysis (Figure S5) revealed that Anita Mahadevan-Jansen, Martin G. Shim, and Eugene Hanlon have significantly influenced the foundational knowledge and methodologies in Raman spectroscopy and tumor research.

Reference

The top 10 references by citation count are presented in Table S3. The top three references (1) Judith Langer et al.(2020, citation counts = 254) reviewed technological advancements of surface-enhanced Raman scattering in the last 45 years; (2) Hyuna Sung et al. (2021, citation counts = 237) highlighted the global cancer burden in 2020 with epidemiological data of various types of cancer, which was frequently cited as background by cancer researchers; (3) Cheng Zong et al. (2018, citation counts = 198) outlined the current advancements in Raman spectroscopy technology and discusses its promising applications in the field of biology. However, the top three references are all reviews. If only considering original articles, the following three studies exhibited the highest citation (1) Michael Jermyn et al. (2015, citation counts = 128) introduced a contact Raman spectroscopy probe for intraoperative detection of invasive brain cancer cells. (2) Holly Butler et al. (2016, citation counts = 128) proposed an experimental protocol for obtaining high-quality Raman images from diverse biological samples. (3) Ximei Qian et al. (2007, citation counts = 113) developed biocompatible, pegylated gold nanoparticles for in vivo tumor targeting. Burst detection analysis (Figure S6) revealed that references such as Abigail S Haka et al. (2005), Ximei Qian et al. (2007), and S Keren et al. (2008) were among the earliest to gain prominence. ,, These works laid the groundwork for the application of Raman spectroscopy in tumor research, particularly in clinical tissues. More recent references, such as Hyunku Shin et al. (2020) and Hollon TC et al. (2020), have emerged as influential, reflecting the growing focus on the combination of Raman techniques and deep learning tools for tumor detection. , The cluster diagram (Figure S7) reveals that the cited references are primarily focused on early pioneering studies applying Raman spectroscopy to cancer tissue diagnosis, the more recent and noninvasive approach of analyzing tumor-derived extracellular vesicles for liquid biopsy, and relevant developments in deep learning, a highly active area in the last five years.

Keywords

and Terms

A bibliometric analysis extracted 1352 keywords extracted from publications in Raman spectroscopy and tumor research. The top five high-frequency keywords included Raman spectroscopy (n = 1654), cancer (n = 1356), spectroscopy (n = 1225), nanoparticles (n = 1172), and diagnosis (n = 744). These keywords were organized into seven major coherent clusters (Figure , small clusters were filtered out), validated by robust structural metrics (modularity Q = 0.4034; silhouette S = 0.7273), indicating significant thematic specialization. The five largest clusters were #0 Raman spectroscopy, #1 gold nanoparticle, #2 surface-enhanced Raman, #3 Raman histology, and #4 molecular docking. A burst detection analysis (Figure ) identified 25 keywords with pronounced citation surges, tracing the field’s progression: pioneering Raman spectroscopy on clinical cancer samples gradually transitioned toward surface-enhanced Raman techniques and nanoparticle synthesis on complex cell and tissue models and later pivoted to liquid biopsy, machine learning, and artificial intelligence.

Unlike keywords, terms are extracted by CiteSpace from high-frequency words in the titles and abstracts. Term analysis has the potential to uncover more latent research connections and emerging directions. The cluster visualization (Figure ) highlights the prominent terminology within the application of Raman spectroscopy in oncology (modularity Q = 0.432; silhouette S = 0.7821), particularly underscoring the rapid emergence of machine learning and deep learning. While the burst detection analysis of terms (Figure S8) reflects this same trend, it also reveals emerging, high-impact research directions, such as surface plasmon resonance, graphene oxide, and photodynamic therapy, which were not identified through the analysis of keywords alone.

Category

Clustering

CiteSpace identified 166 research categories in the field of Raman spectroscopy and tumor research, reflecting their multidisciplinary nature. The three categories with the highest number of publications are (1) Chemistry, Analytical (1624 counts), providing the foundational principles and methodologies for improving the sensitivity and specificity of Raman spectroscopy; (2) Nanoscience & Nanotechnology (1531 counts), which focuses on the development of nanomaterials and nanotechnologies for tumor detection and therapy, particularly in SERS; (3) Chemistry, Multidisciplinary (1422 counts), underscoring the chemical methodologies underpinning Raman-based tumor research. The most central categories by betweenness centrality (BC) were nanoscience and nanotechnology (BC = 0.19), biochemistry and molecular biology (BC = 0.18), and oncology (BC = 0.15). It is noteworthy that Computer Science, Interdisciplinary Applications, as an emerging interdisciplinary field, has experienced a burst in the last five years (burstness = 3.47). The cluster map (Figure ) provides a comprehensive visualization of the interdisciplinary landscape surrounding the application of Raman spectroscopy in cancer research, centered on nanoscience and nanotechnology and biochemistry and molecular biology.

Journal

Analysis and Publication Trends

A total of 1388 journals were involved in publishing research on Raman spectroscopy and tumor research. According to Bradford’s law, 20 journals were identified in Zone One, accounting for 35.8% of the total publications. Considering both publication and citation, the leading journals (Table S4) include Analytical Chemistry (322 publications, H-index = 67), Analyst (295 publications, H-index = 56), Spectrochimica Acta Part A-Molecular and Biomolecular Spectroscopy (234 publications, H-index = 32), and Biosensors Bioelectronics (195 publications, H-index = 58).

The dual-map overlay of journals (Figure ) reflects the interdisciplinary distribution and citation trajectory. The citing journals (left) and cited journals (right) are connected by lines of varying colors and thicknesses, indicating citation relationships and strengths. The analysis showed that studies published in Chemistry/Materials/Physics/Molecular Biology/Genetics were frequently cited by Immunology/Medicine/Dentistry/Surgery/Neurology.This suggests that findings from fundamental sciences often translate into clinical applications, fostering interdisciplinary collaborations and innovation.

Discussion

Although the Raman optical effect was discovered nearly a century ago, it was not until the 1980s, following the application of SERS, that Raman technology began to see increasing research applications in the medical field. The explosive growth of Raman technology in oncology research, however, only emerged after the year 2000 (Figures , S2, and S3). Notably, the year 2024 represents a significant milestone in the field of SERS, commemorating five decades since its initial discovery. Therefore, we have selected literature published between 2000 and 2024 as our research subject, employing bibliometric and visualization methods to systematically review the achievements of Raman optical technology in oncology and analyze its development trends.

In recent years, a number of distinguished scientists in this field have published excellent reviews on the medical applications of Raman spectroscopy. ,,−

Whereas some of these reviews cover broader scopes without focusing on the advances in clinical applications, others are narrowly confined to specific scenarios such as exosome or endoscopy. In contrast, our study comprehensively examines Raman technology’s applications across the entire spectrum of oncology research. While systematically focusing on tumor diagnosis, we also evaluate its roles in photodynamic therapy and drug development. More importantly, this study employs objective, data-driven bibliometric methods to avoid author bias, preventing either overestimation or underestimation of research results. The trend analysis based on bibliometric results also demonstrates greater reliability. Data visualization represents another distinctive feature of this study. By Citespace analysis like network clustering and burst detection, hidden research hubs and emerging hotspots could be identified and highlighted.

Overview of the Bibliometric Analysis

Studies applying Raman spectroscopy in oncology demonstrate significant geographical imbalance in terms of research output, with the top three countries (China, the USA, and India) accounting for 60.12% of total publications (Table S1). While China and India show rapid growth in publication volume, the United States retains its dominance when considering citation and international collaboration. These findings underscore the demand to enhance international collaboration with particular emphasis on incorporating clinical data from underdeveloped regions. The analytical results demonstrate that SERS and its various derivatives (substrate, nanoparticle, etc.) constitute the predominant research focus in this field, as evidenced by their frequent occurrence as keywords and terms. Many of the most productive institutions (Chinese Academy of Sciences, Centre National de la Recherche Scientifique et al., Figure ) and authors (Haishan Zeng, Shangyuan Feng et al., Figure ) have contributed to the application of SERS in cancer research. The development of novel materials, particularly nanoparticles such as gold and silver, has consistently boosted the application of SERS over the past 25 years. It is therefore unsurprising that analytical chemistry and nanoscience take center place in our category clustering (Figure ). Furthermore, research advancements in analytical chemistry techniques and nanomaterials are frequently cited in publications and journals within medicine (Figure ).

The bibliometric analysis reveals that Raman spectroscopy exhibits a distinct preference for application in some specific cancer types rather than covering all oncological types. One primary factor is the inherent depth limitation of Raman spectroscopy, which facilitates data acquisition from superficial human tissues (breast cancer, skin cancer, oral cancer, et al.) and the clinically significant tumors characterized by high incidence rates and critical demand for precise surgical margin (brain cancer, lung cancer, prostate cancer, et al.), requiring surgical access to deeper-seated tumors, also representing one of the principal research targets for Raman spectroscopic applications. Clinical human tissue, whether ex vivo specimens (biopsy or resection) and in vivo intraoperative tissues, is an irreplaceable research object for Raman spectroscopy (Figure , Figure ). Studies utilizing clinical specimens, particularly in vivo samples, typically achieve higher citations and exert greater influence on disciplinary advancement. It is also noteworthy that the rapid development of laparoscopy and endoscopy in surgical procedures offers significant potential for real-time in vivo Raman spectroscopy analysis. However, acquiring clinical samples presents significant recruitment of an adequate patient cohort that meets study criteria; informed consent of participants to undergo invasive procedures and provide specimens for research purposes; completion of rigorous ethical review processes. These clinical challenges have compelled researchers to rely on alternative model systems, particularly cell-based modelsa prevalent approach in oncological research. The widespread adoption of cellular models stems from (1) the relative ease of culturing tumor cells in vitro and (2) the establishment of well-characterized tumor cell lines across diverse tumor types. In addition to established cell lines, exfoliated cells from patients or control subjects and cultured cells derived from ex vivo tissues represent powerful tools for research. Beyond cell models, molecular biomarkers and exosomes extracted from patient body fluid have become major study subjects for Raman spectroscopy after 2015 (Figures , S8). They are easier to obtain compared with surgical tissue. And the minimal invasive diagnostic approaches for cancer based on body fluids (blood, urine, sliver, et al.) are termed “liquid biopsy.”

After 2020, we witness the burst of AI, especially deep learning in the application of Raman spectroscopy in tumor research (Figures , , and S8). And researchers contributing to the field of AI + Raman are receiving growing scholarly recognition (Xiaowei Cao, Yudong Lu, Cheng Chen, Jian Ye, et al.). While traditional chemometrics is not generally regarded as a branch of artificial intelligence, the two fields share several unsupervised learning techniques, such as principal component analysis (PCA). , In fact, chemometric methods have been widely employed in Raman spectroscopy analysis long before the rise of AI (Figure S8). Chemometrics focuses on interpretable multivariate approaches for chemical data interpretation, whereas AI represents a broad computational discipline centered on intelligent systems, machine learning, and automated reasoning. As a versatile tool, AI has the potential to facilitate the evolution of Raman spectroscopy across the entire (1) assisting in substrate design of SERS; (2) accelerating Raman spectral data analysis; (3) enabling discovery of novel tumor biomarkers; and (4) establishing Raman-based signatures for tumor prognosis and drug sensitivity. Our team has fully recognized the catalytic significance of AI in optical research. We anticipate that AI will emerge as the next-generation engine driving advancements in Raman spectroscopy for the oncology investigation.

From this bibliometric analysis and its visualization, Raman spectroscopy, especially SERS, has undergone rapid advancement in oncology research over the past 25 years, yielding a series of techniques with the potential to transform conventional clinical practice. , This field remains in its ascendant phase of development. Notably, recent progress in AI is poised to further propel the application of Raman spectroscopy across multiple dimensions of the tumor research.

Notably, CiteSpace and equivalent bibliometric tools primarily quantify text frequency, co-occurrence strength, and citation metrics and cannot evaluate the academic novelty, technological breakthrough value, or expert-recognized frontier status of a research topic. This may create a fundamental discrepancy between academically recognized research frontiers and bibliometrically identifiable frequency-driven hot spots.

Raman Spectroscopy: From

an Optical Phenomenon to Integrated Detection Systems

Raman scattering fundamentally involves quantized energy exchange between photons and molecular vibrations/rotations. Approximately one in a million photons undergo inelastic Stokes lines (lower frequency) emerge when molecules absorb photon energy to reach excited states, while anti-Stokes lines (higher frequency) occur upon energy release to ground states. Characteristic frequency shifts (Raman displacements) precisely correlate with molecular vibrational energy levels, generating unique spectral fingerprints. Raman spectroscopy generates sharp, bond-specific spectral fingerprints arising from discrete molecular vibrational transitionsdistinct from the broad emission envelopes characteristic of fluorescence spectroscopy, which arises from a fundamentally different photophysical process. While the spectral resolution achievable in Raman measurements is instrument-dependent, the narrowness of Raman bands facilitates discrimination of closely overlapping molecular signals. Additionally, while water is not Raman-silent (it does produce inelastic scattering), its Raman cross-section is relatively weak compared to most biological macromolecules, making Raman spectroscopy far more amenable to measurements in aqueous biological environments than IR absorption spectroscopy, where water’s strong absorption bands severely limit the accessible spectral window.

Early Raman spectroscopy faced limitations due to the weak intensity of mercury lamp illumination (signal intensity ∼10^–10^) and detector sensitivity constraints, restricting applications to simple molecular systems. The 1960s laser revolution brought the first technological leap: laser monochromaticity and high power boosted signals by 10^4^- to 10^5^-fold, establishing Raman spectroscopy as a cornerstone for chemical bond analysis. In 1974, Fleischmann’s team observed anomalous signal enhancement from pyridine molecules adsorbed on roughened silver electrodes, initially misattributed to surface area effects. Van Duyne et al. clarified in 1977 that localized surface plasmon resonance (LSPR) electromagnetic enhancement from metallic nanostructures underpinned this phenomenon, formalizing SERS theory with single-molecule sensitivity. Concurrently, nonlinear optical advances enabled coherent anti-Stokes Raman scattering (CARS): Maker’s theoretical framework (1965) achieved experimental validation in 1974, utilizing dual-laser coherence to amplify anti-Stokes signals beyond conventional sensitivity limits. The Fourier-transform Raman (FT-Raman) system addressed fluorescence Chase’s team employed 1064 nm near-infrared lasers with Fourier-transform algorithms in 1986, facilitating analysis of highly fluorescent materials like coal. 1990s confocal microscopy enhanced spatial resolution to submicron levels (0.5 μm), enabling defect localization and single-cell analysis. Xie’s 2008 stimulated Raman scattering (SRS) microscopy utilized dual picosecond lasers for stimulated amplification, achieving millisecond-scale imaging with background suppression, revolutionizing live lipid metabolism tracking. Refined CARS techniques realized three-dimensional (3D) chemical imaging via optimized phase matching, exemplified by brain myelin distribution mapping. The early 21st century saw the emergence of fiber-optic Raman probes, revolutionizing real-time in vivo diagnostics during oncological surgeries, particularly in brain and breast cancer. , The subsequent decade ushered in a transformative era of multimodal integration and precision medicine, with Raman spectroscopy expanding into liquid biopsy applications for biomarkers, exosomes, and circulating tumor cells. This period also witnessed Raman synergistic platforms with mass spectrometry, magnetic resonance imaging (MRI), and fluorescence microscopy for tumor microenvironment investigation, such as immune cell infiltration and metabolic reprogramming. −

Recent advancements bifurcate into dual Nanofabrication enables controllable SERS substrate design (e.g., gold nanostars, DNA origami), while convolutional neural networks (CNNs) empower parts-per-billion (ppb) detection in complex matrices. Portable systems (532/785 nm lasers) integrated with cloud platforms now dominate field applications like narcotics identification and food safety screening. Emerging frontiers include quantum Raman techniques (entangled photon pairs surpassing classical sensitivity limits) and terahertz Raman spectroscopy (probing low-frequency collective modes), offering novel pathways for ultrafast molecular dynamics and weak interaction studies.

Surface-Enhanced

Raman Spectroscopy (SERS)

The exponential growth of Raman spectroscopy applications in medicine has been fundamentally enabled by advantages of SERS, a trend that aligns precisely with the bibliometric findings of this study. Our analysis demonstrates that SERS consistently emerges as the most critical subclass in Raman spectroscopy for oncology, irrespective of whether clustering is performed by institutions, authors, or keywords. Furthermore, SERS-associated terms such as “nanoparticle” and “probe” frequently appear as prominent nodes in cluster visualizations, underscoring their methodological centrality.

While a powerful tool for molecular characterization, Raman scattering exhibits an inherently low efficiency. Typically, the ratio of Raman-scattered photons to the total scattered photons ranges from ≈1 in 10^6^ to 10^8^. This low probability highlights the challenges in signal detection and underscores the need for highly sensitive detection technologies in Raman spectroscopy. SERS amplifies the inherently weak Raman scattering signal by several orders of magnitude (typically 10^6^–10^14^ times) through interactions between molecules and nanostructured metallic surfaces. The enhancement arises from two primary electromagnetic enhancement (EE) and chemical enhancement (CE). The total SERS enhancement factor is often a product of EE and CE, though EE dominates in most practical applications. SERS can work directly in biofluids (blood, urine) due to reduced water interference. SERS with engineered tags/probes (indirect) can outperform conventional immunoassays such as ELISA in sensitivity and stability. And label-free (direct) SERS captures the analytes of interest and leverages the intrinsic vibrational fingerprints of biomolecules, eliminating the need for synthetic tags. Label-free approach not only circumvents extrinsic interference and potential phototoxicity but also largely streamlines sample preparation protocols and reduces processing time. Notably, variations in key experimental parametersincluding substrate type, laser wavelength, and sample preparationcan substantially alter spectral profiles, complicating cross-laboratory comparisons. Standardized analytical protocols are therefore essential for consistent SERS detection of human serum using certain nanoparticles.

The now-canonical surface plasmon resonance mechanism was first proposed by M. Moskovits in 1978 and formally designated as Surface-enhanced Raman spectroscopy (SERS) by Van Duyne in 1979. , The application of SERS to biological systems dates to 1980, when Cotton et al. successfully detected SERS signals from cytochrome c and myoglobin adsorbed onto silver electrodes. However, initial adoption was hindered by inherent limitations including feeble Raman intensities, prolonged data acquisition durations, and fluorescence background interference. A pivotal advancement occurred in 1995 when Natan’s group demonstrated SERS using monodisperse gold and silver nanospheresa seminal work that both inaugurated the nanotechnology era of SERS and established the paradigm of structurally defined substrate engineering. Concurrently, label-free SERS techniques have yielded significant advancementparticularly in the domain of in vivo imaging. Furthermore, advancements in Raman spectroscopy have been propelled by the demand for high spatial resolution and precise single-crystal surface analysis, leading to the development of alternative techniques such as tip-enhanced Raman scattering (TERS) and shell-isolated nanoparticle-enhanced Raman spectroscopy (SHINERS). , In contrast to conventional SERS, these methods maintain surface integrity while providing consistent signal enhancement across a broad range of materials, including noble metals and semiconductors. , The advancement of SERS is further exemplified by its successful integration with innovative lasers and microfluidic chips. While SERS in the visible (400–700 nm) and first near-infrared (NIR-I, 700–900 nm) ranges has been widely used, these spectral regions face several limited tissue penetration, photodamage, and autofluorescence. Consequently, second near-infrared window (NIR-II, 900–1700 nm) has garnered increasing attention, particularly in the field of tumor imaging. NIR-II SERS offers deeper tissue penetration, lower photodamage, and higher signal-to-noise ratio. Notably, NIR-II demonstrates expanded substrate versatility beyond Au/Ag nanoparticles, enabling effective coupling with alternative plasmonic materials (e.g., Cu) and nonmetallic substrates (e.g., graphene). The aforementioned NIR-II SERS properties render it particularly suitable for human tissue detection. To enhance SERS efficiency in aqueous environments, auxiliary optical trapping beams were implemented to spatially manipulate SERS substratesa methodology termed “optical tweezers.” This integrated optical tweezer-Raman platform has enabled single-molecule detection and single-cell analysis. Microfluidic chip, also known as lab-on-a-chip (LoC), is a miniaturized device that manipulates fluids at the microscale (typically 10–500 μm channels) to perform biochemical assays. These chips integrate pumps, valves, and sensors to control fluid flow, enabling high sensitivity and high throughput liquid biopsy for cancer. When integrated with SERS, the LoC-SERS platform offers highly sensitive and automated detection for cancer biomarkers, exosomes, and cells with minimal sample volume.

Diverse Applications of Raman Spectroscopy

Complementing the innovations of SERS, Raman spectroscopy keeps synergizing with traditional clinical imaging methods such as MRI and optical coherence tomography (OCT), combining their anatomical precision with Raman’s biochemical specificity to achieve molecular imaging. , Raman plus MRI/OCT/photoacoustic multimodal imaging has been applied in diagnosing brain tumors, breast cancer, and prostate cancer. , And this multimodal approach is poised for transformative gains in diagnostic accuracy through AI-driven radiomics integration. In the biomedicine, Raman spectroscopy has also been explored to detect diverse pathogens. Raman spectroscopy offers a culture-free paradigm for pathogen identification with high sensitivity, significantly reducing diagnosis time. Progress has been made in differentiating various bacteria, viruses, and Mycoplasma pneumoniae. , Notably, Raman spectroscopy has been applied to detect COVID-19 virus in rapid and sensitive manners. With recent advancements in microfluid chip and AI, Raman spectroscopy holds promise for achieving ultrasensitive, highly specific, and point-of-care pathogen detection. Antibiotic resistance and residues have also been investigated by SERS. And controlled drugs such fentanyl, morphine, and ketamine can also be quantified using SERS.

As demonstrated by our findings, environmental science represents another major application domain for Raman technology (Figure ). ,, Raman spectroscopy has been employed to detect persistent organic pollutants (POPs) in ecosystems, contaminants such as sulfamethoxazole in water, and nutritional components such as lycopene in food matrices. Raman spectroscopy presents significant potential for applications in nutrition and toxicology, given its ability to assess water and crop safetyfactors that are fundamentally tied to public health. Furthermore, its utility extends to elucidating chemical reaction mechanisms, as evidenced by studies on photosynthesis. Using stimulated Raman scattering (SRS) microscopy, Wang et al. discovered that protozoa from major Chinese waterways bioaccumulate small microplastics (<10 μm), albeit at a low cellular prevalence (2–5%). This finding provides direct evidence for a potential entry point of microplastics into aquatic food webs, highlighting a transmission pathway that could pose significant environmental and health risks.

Raman Spectroscopy in Clinical

Oncology: From Molecular Biomarker Detection to Intraoperative Guidance and Beyond

In the subsequent parts of this section, we will systematically examine the research advances of Raman spectroscopy in oncology, following a progression from small molecules to biomacromolecules, cells, and ultimately tissue-level applications.

Current cancer diagnostics rely on three pillarsmolecular biomarkers, radiological imaging, and histopathologyeach constrained by fundamental limitations. Molecular biomarkers at the genetic or protein level frequently present detection challenges due to their low abundance in clinical samples. Conventional imaging (ultrasound/CT/MRI) fails to detect early-stage microscopic lesions, and identifiable suspicious foci often demonstrate inadequate diagnostic specificity. Invasive biopsy and histopathology risk procedural complications such as hemorrhage and infection and delay therapeutic decision-making. Cellular and tissue heterogeneity arises from distinct compositional profiles of amino acids, carbohydrates, lipids, proteins, and nucleic acideach exhibiting characteristic vibrational modes detectable via Raman spectroscopy. Consequently, Raman spectra serve as quantitative molecular fingerprints that encode the sample’s biochemical identity, enabling the evaluation, characterization, and discrimination of cancer. Moreover, neoplastic molecular signatures (e.g., protein overexpression) and dysregulated cellular metabolism often manifest before tumors reach the detection threshold of conventional imaging methods. Consequently, beyond its spectral fingerprint specificity, Raman spectroscopy exhibits substantial sensitivity for early-stage tumor detection. Unlike traditional tissue biopsies, which require aggressive sampling of tumor tissue, liquid biopsy offers safer, repeatable, and more sensitive diagnostic methods. The unparalleled advantages of SERS in liquid environments have positioned it as a transformative technology for liquid biopsy.

Metabolites: Lipids,

Amino Acids, Glycans, and Metabolic Reprogramming

The dominant Raman signatures in the high-frequency region (2800–3200 cm^–1^) arise from lipid CH stretching vibrations, which exhibit the highest scattering intensity among all biomolecular vibrations. Consequently, lipids have become a research hotspot for Raman spectroscopy in oncology. Recent advances employed nuclear-targeted gold nanocubes as SERS probes to monitor the mitotic progression in malignant cells, revealing concomitant biochemical alterations in lipid assemblies. Raman spectroscopic characterization of breast cancer cells has further established Raman-based stratification of malignancy grades through distinct lipid metabolic fingerprints. Notably, lipid upregulation and reprogramming have been observed by Raman spectroscopy across multiple cancer types, including lung cancer, prostate cancer, colorectal cancer, and melanoma. ,, By employing a highly efficient spectral compressor that narrows the femtosecond laser spectrum without significant energy loss, we achieved a hyperspectral SRS microscopy system with an unprecedented spectral resolution of 5.4 cm^–1^. This breakthrough enables the differentiation of eight saturated lipid types (C8:0 to C22:0), paving the way for revealing aberrant lipid metabolism pathways in cancer. Contorno et al. conducted a systematic review of 41 studies that employed various Raman spectroscopy modalities to characterize breast cancer. Their analysis identified aromatic amino acids, particularly tryptophan, phenylalanine, and tyrosine, as the most significant biomarkers for distinguishing cancerous breast tissues from healthy ones. Glycan chains constitute essential components of glycoproteins, mediating intercellular recognition and communication while playing pivotal roles in tumor immunomodulation. Notably, SERS has been successfully applied for in situ imaging of sialic acids on epithelial cell adhesion molecule (EpCAM), a breast cancer-associated surface protein, as well as for monitoring drug-induced alterations in protein-specific glycosylations. Kopec et al. employed SERS to spatially resolve glycogen, glycosaminoglycans, chondroitin sulfate, and heparan sulfate proteoglycans in both normal and cancerous tissues. Their findings demonstrated significant dysregulation in glycan metabolism in breast adenocarcinomas and medulloblastomas. More recently, a dual-SERS encoding strategy was developed for in situ and in vivo assessment of multiplex protein-specific glycosylation in tumors. SERS nanosensors have been successfully employed for the selective detection of key metabolitespyruvate, lactate, ATP, urea, etc.in complex biological matrices with minimal interference. Due to their high sensitivity and spatial resolution, these nanosensors represent a promising tool for real-time monitoring of cancer cell metabolism.

Alterations in the physicochemical parameters of the tumor microenvironment (TME) and metabolic reprogramming of neoplastic cells have emerged as focal research frontiers in contemporary oncology. Bi et al. introduced a rapid SERS method capable of characterizing metabolic reprogramming patterns in both cell culture media and human serum samples within 15 min. The researchers proposed the term “SERSome” to conceptually describe the ensemble of molecular structures and their functional characteristics in biofluids. Validation studies demonstrated that SERSome-based analysis achieved diagnostic accuracies of 80.8% in the internal test cohort and 73% in the external validation cohort for prostate cancer. Additionally, Raman spectroscopy has been utilized to investigate cancer cell metabolism and responses to anticancer drugs. This technique has been applied to track metabolic reprogramming in KRAS mutant versus wild-type colorectal cancer cells, revealing distinct biochemical pathways associated with oncogenic mutations. Valera et al. developed a SERS platform to characterize metabolic secretome dynamics in 3D pancreatic tumor models. Their findings establish SERS as a powerful analytical tool for optimizing cancer spheroid cultures, particularly through quantitative assessment of tryptophan metabolism. The tumor microenvironment (TME) refers to the cellular and noncellular ecosystem surrounding a tumor, which includes cellular components (e.g., cancer cells, immune cells, stromal cells) and noncellular components (e.g., extracellular matrix, cytokines). The TME is not just a passive scaffoldit actively shapes cancer behavior, treatment response, and patient outcomes. Valera et al. developed a SERS-based analytical platform to characterize purine metabolite secretion in methylthioadenosine phosphorylase (MTAP)-deficient tumor cellsa genomic alteration strongly correlated with adverse clinical outcomes across multiple malignancies. This work elucidates a previously unrecognized tumor-stroma signaling axis that drives microenvironmental reprogramming in MTAP-deleted cancers, providing mechanistic insights into their aggressive pathophysiology. Another study employed pH-responsive molecules as a Raman reporter probe to functionalize immune-SERS nanotags, enabling the simultaneous identification of extracellular microenvironmental pH and VEGF assay. This approach facilitates the elucidation of microenvironmental impacts on cellular cytokines, which is of paramount importance for early cancer diagnosis and prognosis.

Nucleic Acids

Dysregulation of nucleic acidsvia mutations, epigenetic changes, or aberrant signalingdrives uncontrolled cell growth, immune escape, and metastasis. Raman spectroscopic detection of genes presents inherent challenges due to the limited molecular variability among nucleic acids, which are fundamentally composed of four nucleobases. To address this limitation, advanced methodologies employing dye-coded nanoparticles, microfluidic platforms, and novel materials such as graphene have been developed for enhanced genomic analysis. Pioneering work in Raman Spectroscopy for tumor-derived nucleic acids Abell et al. introduced a highly reproducible method with silver nanorod substrates to determinate micro-RNA sequences; Lu et al. developed a method by linear decomposition of stimulated Raman scattering images to detect DNA from background noises of proteins; El-Sayed et al. demonstrated that DNA conformational differences between healthy and cancerous cells could serve as valuable diagnostic indicators. Recent advances Lemoine et al. conducted a systematic investigation incorporating 547 in situ Raman spectra acquired from 65 glioma patients undergoing tumor resection, supplemented by comprehensive literature meta-analysis. Through advanced band fitting analysis of Raman spectral features, the researchers identified elevated nucleic acid content in neoplastic tissues. In human saliva samples, Liu et al. developed an innovative SERS platform that combines nicking endonuclease-assisted target recycling signal amplification with electrothermal microelectrodes for oral cancer-associated DNA detection. Ma et al. developed an integrated microfluidic SERS platform incorporating DNA cascade signal amplification. This analytical system achieves detection sensitivity of 10.9 particles/μL with a rapid assay time of 35 min, showing promise of point-of-care diagnostic applications.

DNA methylation is an epigenetic modification where a methyl group is added to the cytosine base in DNA. In cancer, DNA methylation patterns are aberrantly altered, contributing to tumor initiation, progression, and metastasis. Using SERS methods which leave the DNA structure intact can have widespread applications in high-throughput screening which can obtain not only genetic but epigenetic information as well. Singh et al. developed an innovative approach integrating magnetic separation with SERS for the selective enrichment and detection of methylated tumor suppressor gene promoters. MicroRNAs (miRNAs) are small, noncoding RNA molecules that regulate gene expression post-transcriptionally by binding to complementary sequences in the 3′ untranslated region (UTR) of target mRNAs. miRNAs, particularly miR-21, have emerged as promising biomarkers for early cancer detection. Chheda et al. developed a machine learning-based analytical framework for the classification of SERS spectra derived from short single-stranded DNA and RNA oligonucleotides. This approach enables precise identification of genetic mutations in known cancer biomarkers, including miR-21. Recently, Yue et al. developed a DNA nanomachine capable of achieving tumor cell-specific miRNA imaging. Both 3D and microfluidic technologies have been integrated with SERS to progressively enhance the detection sensitivity of miRNAs. Recent advances have extended SERS-based RNA profiling to include long noncoding RNAs (lncRNAs), with clinical validation for the screening of liver cancer.

Circulating tumor DNAs/RNAs (ctDNA/RNA) are fragments of tumor-derived DNA/RNA released into blood, offering new possibilities for minimally invasive cancer diagnosis. Zhang et al. developed an innovative frequency-shift-based SERS for the quantitative detection of ctDNA in serum. The analytical platform achieved a limit of detection of 0.12 fM, with significant potential for early-stage lung cancer diagnosis. Zheng et al. developed an innovative label-free analytical platform combining SERS with machine learning for molecular profiling of serum DNA. Their approach achieves simultaneous detection of both epigenetic modifications and genetic mutations at single-nucleotide resolution. SERS has also emerged as an ultrasensitive tool to profile ctRNA, particularly circSATB2, screening cancer at the early-stage circSATB2.

Serum

Protein Biomarkers

Clinically established serum diagnostic biomarkers are predominantly protein-based. Protein biomarkers in diagnostics typically exhibit well-established structures and have the corresponding antibodies available for specific recognition. The most prevalent detection methods for protein biomarkersincluding enzyme-linked immunosorbent assay (ELISA) and radioimmunoassay (RIA)are based on the principle of antigen–antibody interaction, constituting what is collectively termed an immunoassay. Raman-based immunoassays constitute a class of analytical techniques that utilize Raman spectral signatures as quantitative readout signals. The inaugural SERS-based immunoassay was established in 1999. In this pioneering work, Raman reporter-labeled gold nanoparticles were conjugated with detection antibodies for an immunoassay.

Prostate cancer represents one of the most prevalent malignancies in males, with PSA serving as the primary diagnostic biomarker. Consequently, PSA has become one of the most well-established protein biomarkers and is frequently employed to validate novel Raman-based immunoassays. As early as 2003, Grubisha et al. developed gold nanoparticles (30 nm) for a SERS-based immunoassay to detect PSA in human serum and achieved a limit of approximately 1 pg/mL with a readout time of 60s. Chen et al. recently developed a double-SERS satellite immunoassay utilizing Au–Ag nanoparticles, establishing an innovative PSA-mediated Prostate Health Index diagnostic platform. This method demonstrates enhanced predictive accuracy and specificity for prostate cancer, particularly in the “diagnostic gray zone,” where PSA falls within the diagnostically ambiguous 4–10 ng/mL range. Hepatocellular carcinoma represents a more aggressive and lethal malignancy compared to prostate cancer, with AFP serving as its most established diagnostic biomarker. Chen et al. developed novel Mo2N nanoparticles with both enzyme-like and SERS sensitivity achieving an ultrasensitive AFP detection limit of 89.1 pg/mL.

Beyond enhancing detection sensitivity, another critical research direction for Raman-based protein biomarker analysis focuses on achieving high-throughput, rapid, and multiplexed detection capabilities. For colorectal cancer diagnosis, Cao et al. integrated a microfluidic chip with SERS to construct a multiplex profiling platform, enabling to detect multiple tumor markers in high-throughput samples. This method has achieved LODs of 0.057 pg/mL (S100P) and 0.031 pg/mL (hnRNP A1) within 15 min. Seo et al. developed a novel SERS immunoassay platform for rapid quantification of multiple cancer biomarkers. Validation studies demonstrated the assay’s high diagnostic performance, achieving 92% accuracy in lung cancer detection using 20 μL of blood serum. The platform exhibited robust classification capabilities, correctly identifying cancer subtypes and staging with accuracies of 87% and 85%, respectively.

Regarding serum biomarkers, recent studies have identified uric acid and hypoxanthine as the main contributors to the SERS spectra of biofluids such as serum and plasma when measured using Ag substrates under near-infrared excitation. Furthermore, isotopic labeling experiments confirmed that both albumin-bound and free uric acid, which display distinct spectral features, contribute to the overall serum SERS signal. These findings clarify the biochemical origin of the dominant bands observed in serum SERS spectra and provide a foundation for future investigations. ,

Extracellular

Vesicles and Exosomes

Extracellular vesicles (EVs) are lipid bilayer-enclosed particles released by virtually all cell types into the extracellular space and bodily fluids. They play critical roles in cell-to-cell communication and cancer progression. Exosomes are a subtype of EVs with a diameter of 30–150 nm. As natural carriers of tumor-derived metabolites, nucleic acids, and proteins, exosomes have emerged as promising liquid biopsy targets, with their lipid bilayer ensuring cargo stability in circulation. , A key advantage of Raman spectroscopy, particularly SERS, lies in its capacity for rapid, accurate, and high-throughput detection of biomarkers in bodily fluids. Consequently, one major research direction in tumor-derived exosome analysis focuses on optimizing Raman platforms to enhance detection efficiency for exosomal metabolites, nucleic acids, and proteins. , Zhao et al. developed a novel CD9-labeled SERS platform to detect exosomes in urine samples, revealing significantly enhanced SERS intensity from pancreatic cancer patients compared to healthy controls. Chen et al. developed a SERS-based multichannel microfluidic platform for EV phenotyping analysis. The assay demonstrated robust diagnostic performance for early-stage (I–II) ovarian cancer detection, exhibiting an area under the curve of 0.947.

Another distinctive capability of Raman technology is its ability to interrogate exosomes as “holistic spectroscopic units,” enabling biomarker-free early cancer diagnosis through comparative analysis of intrinsic Raman spectral variations of exosomes across clinical samples. Carmicheal et al. demonstrated the capability of label-free SERS to discriminate between normal and malignant exosomes through Raman spectra. This method achieved 90% diagnostic accuracy in differentiating pancreatic cancer patients from healthy controls.

The advancement of AI has significantly augmented the analysis of Raman spectroscopic data, positioning exosome-based biomarker-free liquid biopsy as a frontier in cancer diagnostics. Xie et al. presented an AI-SERS platform for the label-free spectroscopic characterization of serum-derived exosomes. This innovative approach enabled both accurate breast cancer diagnosis and quantitative evaluation of surgical treatment efficacy. The developed deep learning algorithm, trained on SERS spectral profiles of cancer-cell-derived exosomes, achieved ideal diagnostic discrimination (100% accuracy) in classifying diverse breast cancer subtypes. Premachandran et al. established a SERS-based machine learning algorithm trained on EV molecular signatures, achieving 97% sensitivity in discriminating metastatic from primary brain malignancies.

Circulating Tumor Cells

and Cellular Models

Cells are the basic structural and functional unit of living organisms. Circulating tumor cells (CTCs) are cancer cells that detach from primary or metastatic tumors and enter the circulation system. CTCs can be difficult to detect as there are typically just 1–10 per milliliter of blood in cancer patients. Raman spectroscopy, with its unparalleled single-cell resolution and compatibility with other imaging technologies, can bridge the gap between molecular specificity and clinical practicality, making it a transformative tool for CTC research.

Pioneering work by Sha et al. in 2008 demonstrated SERS nanotag labeling of CTCs through biomarker-targeted surface functionalization, coupled with magnetic bead enrichment for quantitative SERS analysis. Cellular phenotypes are characterized by distinct surface protein expression profiles, which can serve as effective molecular targets for both cell sorting and Raman spectroscopic imaging. Biomarker-free approaches for cellular phenotyping also emerged through the development of Raman spectroscopy, which bypasses conventional fluorescent labeling by directly probing the intracellular molecular composition via Raman spectroscopic measurements. Recent advances incorporating coherent Raman scattering modalitiesparticularly stimulated Raman scattering (SRS) and coherent anti-Stokes Raman scattering (CARS)have significantly augmented light–matter interactions, thereby facilitating high-throughput cancer cell detection.

Advances in 3D cell culture models and microfluidic chip technologies have significantly expanded the applications of Raman spectroscopy in cancer cell research. As emphasized by Lin et al., the emerging application of SERS for 3D biomodel imaging necessitates rigorous validation of its spatiotemporal correlation with confocal fluorescence microscopy. González-Callejo et al. developed an advanced 3D bioprinted tumor-stroma model incorporating triple-negative breast cancer stem cells and human mammary fibroblasts within a decellularized breast extracellular matrix bioink. This biomimetic platform enabled SERS-based spatial mapping of tumor cell invasiveness across deep z-axis planes, circumventing the penetration limitations of conventional confocal fluorescence microscopy. Yang et al. developed a dynamic liquid-integrated single-cell analysis platform combining a twisted mixing microfluidic chip with SERS for label-free cancer cell detection. The system demonstrated exceptional classification performance, achieving accuracy, sensitivity, and specificity up to 99.5%, 100%, and 99.4%, respectively. For living cells, our team has developed a minimally invasive Raman reporter (<1 kDa, ∼2 amino acids) that is site-specifically incorporated into proteins via genetic codon expansion, enabling bioorthogonal labeling through a copper-free click reaction with a tetrazine-functionalized Raman tag. This strategy allowed precise, multicolor stimulated Raman imaging of vimentin, histone 3.3, and huntingtin in living HeLa cells, establishing a powerful platform for minimally invasive protein tracking and multiplexed live-cell imaging. With continued advancements in Raman-based CTC researchparticularly through AI integrationSERS has achieved unprecedented detection sensitivity of 1–2 cells/mL while maintaining exceptional classification accuracy. ,

Tissue Discrimination and Surgical Guidance

Tissue refers to an organized assembly of cells and extracellular matrix that performs a specific function in an organism. Raman spectroscopy offers unique capabilities for real-time tissue analysis with high spatial resolution while preserving tissue integrity. Since pathological examination is the “gold standard” for cancer diagnosis, which relies on biopsied or surgically resected human tissues, Raman spectroscopic analysis of tissue specimens has remained a central focus in this field. Accurate surgical marginthe boundary between resected tumor tissue and surrounding healthy tissueis a cornerstone of successful cancer treatment. Incomplete resection leaves behind microscopic tumor cells, leading to cancer recurrence. On the opposite, over-resection will harm critical anatomical and functional structurea scenario with potentially catastrophic consequences in intracranial procedures. Thus, Raman spectroscopic tissue discrimination serves not merely to enhance conventional histopathological accuracy but more significantly as a promising intraoperative guidance tool for surgical optimization. Our bibliometric analysis of keywords and terms frequency networks consistently identified “tissue” as the predominant cluster, underscoring its primary research emphasis (Figure ).

Breast cancer, the most prevalent malignancy in women, represents one of the most extensive applications of Raman spectroscopy in tissue analysis. As early as 1993, Redd et al. pioneered this approach by characterizing Raman spectral differences between normal and neoplastic breast tissues, demonstrating diagnostic potential through distinct heme-type signals and lipid contributions. Between 2000 and 2010, Haka and colleagues conducted a seminal series of Raman spectroscopic studies on breast cancer tissues. , In 2006, Haka et al. pioneered the first in vivo Raman spectroscopy for breast tissue, capturing 31 intraoperative spectra. Established modeling enabled <1 s tissue classification, suggesting clinical utility for margin assessment and re-excision reduction. Their innovative methodology utilizing microcalcification Raman signatures for tissue discrimination has profoundly influenced approaches of other studies. Our bibliometric analysis identifies their publications as the most-cited works in this domain. Since 2020, Raman spectroscopy in breast cancer tissue research has evolved toward multimodal integration and diversified clinical scenarios. Ni et al. developed a label-free Raman histology platform and successfully generated spatially resolved maps of key breast tissue components including duct, necrosis, and vessel.

The intracranial compartment houses intricate anatomical structures that serve as centers for language and motor functions. Precise discrimination between neoplastic and benign intracranial tissues is therefore clinically paramount. Malignant intracranial tumors represent both a challenging and compelling research target for Raman spectroscopic applications. Jermyn et al. developed a hand-held contact Raman probe for intraoperative detection of live glioma cells at cellular resolution. The device accurately discriminated between normal brain tissue, solid tumor (sensitivity: 93%; 91%), and diffusely invasive cancer cells in grade 2–4 gliomaaddressing the limitation in visualizing microscopic tumor infiltration. By our statistics, this original work has been cited most often as a reference of studies in the field of Raman spectroscopy in oncology. Recently, Zhu et al. established Raman spectroscopy as an effective tool for detecting glioblastoma microinfiltration at cellular resolution in clinical tissue. When combined with machine learning, the technique achieved >95% AUC in identifying infiltrative lesions. Notably, the detection threshold reached 3 tumor cells/0.01 mm^2^, demonstrating an exceptional sensitivity for minimal residual disease.

The prostate gland serves as a critical organ that is responsible for both urinary and sexual functions in males. Accurate discrimination between cancerous and benign prostatic tissues carries profound implications, not only for surgical outcomes but also for long-term quality of life preservation. With AI integration, Mannas et al. have significantly advanced label-free SERS methodology for prostate tissue discrimination, achieving exceptional performance metrics of 96.3% sensitivity and 96.6% specificity. In a subsequent clinical study, the same team implemented stimulated Raman histology during robotic-assisted laparoscopic prostatectomies in 22 patients. The team demonstrated 98% accuracy in interpreting resected periprostatic surgical bed tissue using stimulated Raman histology with 83% sensitivity and 99% specificity. Intraoperative stimulated Raman histology assessment identified 43% of the participants with pathologically positive surgical margins during the procedure.

Beyond its applications in high-incidence malignancies such as breast and prostate cancers, Raman spectroscopy is increasingly employed for superficial tumor tissue analysisparticularly in skin and oral carcinomaowing to its technical advantages for accessible tissue spectroscopy. Our aforementioned cluster analysis substantiates these research paradigms.

AI integration will undoubtedly enhance the discriminatory capacity of Raman spectroscopy for tumor tissue characterization, offering transformative potential for precision surgery. Reinecke et al. have pioneered AI-augmented stimulated Raman histology (SRH) for intracranial lesions, achieving unprecedented intraoperative virtual pathology diagnostic accuracy of 97.8% with rapid processing under 3 min. Their large-scale, multicenter prospective external validation studies establish this approach as a promising paradigm for enhancing surgical decision-making efficiency and improving patient outcomes.

The relatively limited application of Raman spectroscopy in living human tissues primarily stems from the stringent ethical approvals required for in vivo human studies. Since most clinically prevalent tumors have well-established animal analogs, the majority of Raman-based live tissue investigations have been validated using animal modelsparticularly xenograft tumor models in immunocompromised mice. Qian et al. in 2008 demonstrated the first in vivo SERS detection of tumors using EGFR-targeted Au-nanotags, validating molecular Raman imaging in cell cultures and murine xenografts. Subsequently, Raman spectroscopy has been extensively employed in live animal models of breast cancer, glioblastoma, prostate cancer, et al. , Beyond tumor tissue analysis, Raman techniques have enabled precise discrimination of metastatic sentinel lymph nodes, blood vessels, and lymphatic channels in a preclinical modelproviding novel intraoperative guidance for optimizing surgical strategy.

Endoscopy

Endoscopy is a minimally invasive procedure using a camera-equipped scope to visualize and treat internal organs. Endoscopy plays a pivotal role in modern oncology by enabling early tumor detection and targeted biopsies. For early-stage tumors, it also offers curative treatments such as endoscopic mucosal resection of small gastric cancer and transurethral resection of bladder tumor. Gastroscopes, colonoscopes, cystoscopes, and bronchoscopes represent the most widely utilized clinical endoscopes and, consequently, prime candidates for Raman spectroscopy integration in endoscopic platforms. During the 2000–2010 period, Xie et al. pioneered coherent anti-Stokes Raman scattering (CARS) endoscopy investigations, leveraging the technique’s inherent label-free and chemical selectivity for in situ endoscopic imaging. In clinical studies, Raman spectroscopy achieved differential diagnosis of colorectal and bladder tumors from endoscopy with high accuracy. Employing machine learning, Jo et al. achieved perfect sensitivity (100%), high specificity (93.3%), and excellent accuracy (96.7%) in discriminating colorectal cancer from normal murine tissues.

Anticancer

Drugs

While tumor diagnostics remains a predominant research focus for Raman spectroscopy in clinical medicine, our bibliometric analysis reveals emerging emphasis on drugs. The intracellular dynamics and distribution mechanisms of many clinically established anticancer agents remain incompletely characterized, impeding research on drug resistance and efficacy enhancement. Raman spectroscopy offers unique advantages for elucidating drug uptake processes, as evidenced by extensive investigations in this domain. Raman spectroscopy’s exceptional sensitivity also enables trace detection of therapeutic agentincluding antineoplastics, antibiotics, and antidepressantin patient biofluids such as blood and urine. As a highly versatile analytical platform, Raman spectroscopy has been successfully integrated with mass spectrometry and metabolomics approaches, enabling novel anticancer drug targets discovery. Raman spectroscopic analysis of patient-derived biofluids has enabled identification of drug-response spectral signatures, with demonstrated applications in predicting therapeutic sensitivity for prostate and breast carcinomas.

Cancer immunotherapy is transformative in modern oncology, offering durable responses in previously untreatable cancers and revolutionizing combination therapies. Immunotherapy’s significance was recognized with the 2018 Nobel Prize in Physiology or Medicine. Immunotherapy harnesses the immune system to target cancer cells through mechanisms such as immune checkpoint inhibition (e.g., PD-1/PD-L1, CTLA-4 blockade), enhancing T-cell-mediated tumor destruction. Li et al. established novel frequency modulation protocols enabling 32-plex discrimination in the Raman-silent region. And by this capacity, the researchers could predict combined drugs of immune checkpoint inhibitors based on SERS. Paidi et al. pioneered the use of label-free Raman spectroscopy to characterize tumor microenvironment biomolecular alterations following anti-CTLA4 and anti-PD-L1 immunotherapy in colorectal xenografts. Zhou et al. developed machine learning cascade (MLC)-enhanced Raman histopathology to spatially resolve PD-L1 expression across glioma cells, CD8+ T cells, macrophages, and normal parenchyma while delineating tumor margins.

Photodynamic

Therapy and Beyond

Our terms analysis identifies photodynamic therapy (PDT) as a prominent research focus (Figure S8), demonstrating its role as one of Raman spectroscopy’s primary applications in oncological therapeutics. PDT is a minimally invasive cancer treatment that utilizes light-absorbing agents (e.g., gold nanoparticles) to convert light into localized heat, inducing tumor cell death. Raman spectroscopy can be employed in PDT to guide treatment precision by mapping tumor-specific molecular signatures for targeted laser irradiation, monitor thermal effects in real-time, and verify agent distribution with high spatial resolution. And label-free SERS can complement PDT by enabling pretherapeutic diagnosis, intraoperative margin assessment, and post-treatment evaluation of cellular damage. Liu et al. developed a theranostic nanoplatform based on gold nanorods functionalized with manganese porphyrins, enabling dual-modal SERS/photoacoustic imaging-guided PDT. Ex vivo and in vivo studies confirmed its efficacy in precise tumor localization and ablation. Li et al. then developed a NIR-II-responsive plasmonic immunomodulator that synergized photothermal therapy with an αPD-1 blockade. This integrated platform enables SERS-guided combination photoimmunotherapy, demonstrating high therapeutic efficacy through hyperthermia-triggered immunomodulation.

Contemporary medicine has expanded beyond tumor diagnosis and treatment to emphasize proactive health management. The advancement of wearable electronic devices now enables continuous, out-of-hospital health monitoring. Emerging research demonstrates successful integration of Raman spectroscopy with wearable platforms for tracking glucose and sweat biomarkers.

Advances in Raman Spectroscopy for Oncology

Research: From Nano-Driven to AI-Driven

The remarkable progress in Raman spectroscopy, particularly in SERS, has been fundamentally enabled by advancements in nanoscience. This conclusion is supported by our bibliometric analysis, where multiple nanoscience-related keywords and terms (e.g., gold nanoparticles, carbon nanotubes, nanocrystals) occupy central positions in cluster analysis (Figure ). Moreover, in the field of Raman spectroscopy for tumor research, the most productive institutions (Figure ) and authors (Figure ) have consistently utilized nanotechnology to advance their Raman spectroscopic investigations.

Continued

Evolution of Nanoscience

As the cornerstone of SERS biosensors, substrate design governs critical analytical parameters, including sensitivity, reproducibility, and detection accuracy. Nanostructures serve as fundamental components for both label-free and indirect SERS detection, as their architectures and optical properties directly govern detection sensitivity. Gold and silver nanoparticles remain predominant SERS substrates due to their facile synthesis, reproducible enhancement, and strong plasmonic response. Significant research efforts have focused on optimizing SERS enhancement through engineered core–shell Au and Ag nanostructures, improving Raman signals and substrate chemical stability. Concurrently, antibody-conjugated metal nanoparticles targeting specific tumor biomarkers (e.g., EGFR, HER2) have emerged as precision tools for cancer diagnostics. Recent advances in SERS substrate design have leveraged novel nanomaterials such as semiconductors, transition-metal dichalcogenides, and metal–organic frameworks to enhance sensitivity, stability, and reproducibility. These materials offer tunable plasmonic properties and high surface areas, significantly improving the SERS performance compared to conventional metallic substrates. Graphene is a two-dimensional (2D) nanomaterial composed of a single layer of carbon atoms arranged in a hexagonal lattice. Carbon nanotubes are cylindrical nanostructures composed of rolled-up graphene sheets. These novel nanomaterials exhibit exceptional properties for Raman spectroscopy, including high electrical/thermal conductivity, mechanical strength, and large surface area. Three-dimensional (3D) printing also emerged as a versatile substrate fabrication technique, enabling computer-controlled assembly of materials into tailored 3D architectures. Nguyen et al. developed a flexible 3D plasmonic cluster SERS platform using bottom-up assembly for early breast cancer detection. When integrated with deep-learning analysis, the system achieved 93% accuracy in discriminating cancer patients from healthy controls based on spectral signatures.

Transformative Emergence of AI

Recent advances in AI have revolutionized Raman spectroscopy, propelling it into a new era of precision and efficiency. , As demonstrated in our previous study, by leveraging deep learning architecturessuch as convolutional neural networkresearchers can now decode intricate spectral patterns with unprecedented accuracy, achieving diagnostic performance that surpasses traditional analytical methods. These AI-driven approaches not only improve classification reliability but also facilitate the discovery of previously unrecognized molecular signatures, thereby expanding the technique’s utility in biomarker research and clinical diagnostics.

AI is the broad field of enabling machines to perform tasks requiring human-like reasoning. Machine learning, a subset of AI, uses statistical algorithms to learn patterns from data without explicit programming, while deep learning, a specialized machine learning branch, employs multilayered neural networks to model complex, hierarchical data representations. ,, In Raman spectroscopy for cancer research, machine learning (e.g., support vector machine or SVM, principal component analysis or PCA) classifies spectral data by distinguishing tumor/normal tissues based on preselected features, whereas deep learning (e.g., convolutional neural networks or CNNs) autonomously extracts high-dimensional features from raw spectra, enabling precise tumor subtyping or grade prediction. While machine learning typically relies on handcrafted features and performs well on small-to-moderate data sets, it was widely employed long before the advent of deep learning. Our analysis shows that machine-learning-related publications began to surge in 2011 and still account for a larger body of literature than deep learning (Figures , S8). In contrast, deep learning has emerged as a rapidly growing topic in Raman spectroscopy in 2020. AI integrates these tools to automate diagnosis, enhance signal interpretation, and guide personalized therapies, as demonstrated in SERS-based liquid biopsies.

A growing number of sophisticated reviews has comprehensively documented AI’s transformative impact across various aspects of Raman spectroscopy, cataloging its technological innovations and clinical applications. , Our bibliometric analysis of keywords, terms, and citation rates reveals that AI-related words have surpassed nanomaterials as the predominant research focus in Raman spectroscopy applications for tumor studies over the past five years. A marked surge in AI-related keywords and terms, particularly “deep learning,” has been observed in Raman spectroscopy applications for cancer research since 2020 (Figures , S8). Although these AI-related keywords and terms emerged later than nanoscience ones in the timeline analysis, computer science has formed a distinct research category (Figure ). Notably, the most frequently cited original articles in this field over the past five years predominantly incorporate AI-based methodologies. A representative example is the 2020 study by Hollon et al., which presented an automated parallel workflow integrating stimulated Raman histology (SRH) with deep CNNs for near real-time intraoperative diagnosis. The CNNs, trained on >2.5 million SRH images, deliver accurate tumor classification in <150 s10 × faster than conventional histopathology (20–30 min).

As previously discussed, AI-augmented Raman spectroscopy has demonstrated progressive improvements in accuracy across multiple oncological research domains from molecular profiling and exosome characterization to tissue diagnostics. Here are more recent examples highlighting advancements in Raman spectroscopy with AI integration. Yang et al. developed an integrated microfluidic-SERS platform coupled with machine learning for automated classification of hematologic malignancies. The system distinguished acute lymphoblastic T-cell leukemia from chronic myeloid leukemia with 98.6% accuracy in clinical validation (n = 73). Chen et al. developed an integrated microfluidic-SERS platform incorporating deep learning for high-precision exosome analysis in nonsmall cell lung cancer (NSCLC). The system (1) 85% exosome capture efficiency through optimized microfluidic enrichment and (2) 97.88% classification accuracy with AUC >0.95 in discriminating three NSCLC subtypes (adenocarcinoma, squamous cell carcinoma, large cell carcinoma) from normal bronchial epithelial cells, based on exosomal surface protein profiles. Hu et al. developed an innovative self-supervised learning framework for Raman spectral preprocessing that preserves exceptional spectral fidelity while enabling large-scale medical analysis. Zhang et al. developed a Ramanome-based Metastasis Index (RMI) leveraging machine learning analysis of single-cell Raman spectra for rapid, quantitative assessment of metastatic potential. AI augmentation also expands Raman spectroscopy’s diagnostic paradigm beyond conventional tumor detection approaches. Shin et al. demonstrated an AI-enhanced SERS platform for simultaneous multicancer early detection through exosome profiling, evaluating six cancer types (lung, breast, colon, liver, pancreas, stomach) in a retrospective study. The system achieved an AUC of 0.970 for cancer detection (n = 520 samples) and 0.945 mean AUC for tumor origin classification (n = 278 early-stage patients), showcasing high diagnostic accuracy for both malignancy identification and tissue-of-origin determination. Calvarese et al. engineered a multimodal nonlinear optical endomicroscope integrating coherent anti-Stokes Raman scattering, two-photon fluorescence, and second-harmonic generation with deep learning algorithms for label-free tissue analysis. This system achieved 88% sensitivity and 96% specificity in early head and neck cancer detection. The platform also incorporated machine learning-guided femtosecond laser ablation for targeted therapeutic intervention, enabling real-time “seek-and-treat” functionality. Furthermore, AI-Raman enables rapid processing of large-scale clinical data sets while reducing reliance on traditional biopsy and streamlining conventional workflows, thereby facilitating point-of-care tumor diagnostics.

Dual Engines: AI + Nano

It should be emphasized that “AI-driven” does not imply diminished importance or replaceability of material research but rather optimizes key processes including nanomaterial design, screening, and validation, thereby significantly enhancing the efficiency of novel nanomaterial development. Notable examples include the work by Fang et al., who developed machine learning-optimized SERS substrates for detecting lung cancer biomarkers in complex exhaled breath. Furthermore, critical nanomaterial concepts like “molecular docking” have been significantly advanced through AI integration. Molecular docking has emerged as a prominent term in Raman spectroscopic oncology over the past five years (Figure , Figure ). This computational approach simulates intermolecular interactions to predict binding conformations and affinity between molecules (e.g., Raman-active nanoparticles and their biological targets), finding critical applications in SERS substrate design. Although fundamentally a structural chemistry concept, its rapid adoption in Raman spectroscopy has been significantly accelerated by advances in AI.

We highlight AI’s emergence as a powerful auxiliary tool that complements, rather than replaces, materials science in this field. AI bridges the gap between Raman’s molecular specificity and clinical scalability, enabling faster, cheaper, and more precise cancer diagnostics and therapeutics.

Challenges and Future Directions

Driven successively by nanoscience and AI, Raman spectroscopy has achieved significant preclinical advances in tumor diagnosis and therapy. Based on a comprehensive review of its 50 year application in oncology research, particularly quantitative bibliometric analyses from the past 25 years, we posit that Raman spectroscopy’s development in this field remains in an exponential growth phase, far from reaching a research plateau. However, this does not imply that Raman-based technologies for clinical cancer diagnosis and treatment have reached maturity. We argue that for Raman spectroscopy to become a transformative technology in clinical practice, several critical challenges must be addressed.

Intrinsic Technical Limitations of Raman

Spectroscopy

Raman spectroscopy faces several intrinsic challenges that constrain its broader adoption in medical applications. First, the inherently weak Raman scattering effect, with cross sections about 10^6^ and 10^14^ times lower than infrared absorption or fluorescence, necessitates signal-enhancement strategies such as SERS. While SERS improves sensitivity, it introduces new limitations, including nanostructure instability and poor signal reproducibility due to inconsistent “hot spot” formation. Quantitative analysis remains challenging because of heterogeneous signal enhancement and susceptibility to environmental perturbations. ,

Beyond substrate optimization, recent research has investigated innovative illumination approaches to improve the tissue penetration depth of Raman spectroscopy. Zhou et al. demonstrated plasmonic surface-enhanced transmission Raman spectroscopy (SETRS) for deep-tissue lesion detection, achieving unprecedented 2.6 cm penetration depth in live rat models. NIR-II wavelength irradiation and single-photon technologies have emerged as promising solutions. Our team has conducted systematic research on enhancing medical optical imaging quality, particularly in single-photon technologies and artificial intelligence. ,−

These optical advancements have been applied to cancer research, improving the accuracy and reliability of single-molecule imaging within cancer cells. Regarding tissue penetration depth, we developed a high-frequency near-infrared up-conversion single-photon imaging platform. Based on quantum compressed sensing, this platform leverages the single-photon characteristics in the NIR-II to significantly improve signal acquisition efficiency in deep tissues. Single-photon imaging techniques, utilizing principles such as quantum coherent modulation and compressed sensing, significantly improve signal-to-noise ratio and imaging resolution. We developed a novel Quantum Coherent Modulation-Enhanced Single-Molecule Imaging Microscopy, achieving ultrahigh sensitivity detection of single-molecule signals in complex biological samples by modulating the interaction between single photons and molecules. More recently, we introduced the fiber-array Raman engine (FIRE), a platform enabling nonrepetitive, single-shot spectral acquisition at MHz rates across the full Raman span (−300–4300 cm^–1^) with superior sensitivity. This technology, demonstrated through metabolic imaging of intact C. elegans, outperforms conventional Raman microscopes in speed, spectral coverage, and autofluorescence suppression, establishing a new paradigm for real-time biochemical analysis.

Coherent Raman scattering (CRS) approaches, including coherent anti-Stokes Raman scattering (CARS) , and stimulated Raman scattering (SRS), offer an important class of solutions to the weak signal limitation intrinsic to spontaneous Raman spectroscopy. By exploiting the coherent amplification of molecular vibrations using synchronized dual-laser pulses, CRS techniques can achieve signal intensities orders of magnitude greater than spontaneous Raman, enabling high-speed, label-free chemical imaging of living cells and tissues at video-rate acquisition. SRS in particular offers the advantage of a background-free signal that scales linearly with analyte concentration, making it compatible with quantitative imaging. While CRS methods circumvent the signal intensity bottleneck, they introduce their own technical requirements, including the need for precisely synchronized ultrafast laser pulses, increased system complexity, and susceptibility to nonresonant background in CARS. Nonetheless, their integration with advanced optics and AI-assisted spectral analysis holds considerable promise for expanding Raman spectroscopy’s clinical applicability.

The high cost of advanced Raman systems, particularly SERS-integrated endoscopic platforms, limits accessibility in low-resource settings, exacerbating disparities in cancer diagnosis. This study reveals a significant geographical imbalance in Raman spectroscopy advancements with concentrated outputs primarily localized in East Asia, North America, and Europe. The integration of AI and novel materials is expected to reduce both the development and operational costs of Raman spectroscopy systems.

Stringent Requirements

for Clinical Applications

The unique characteristics of biological samples further complicate the Raman spectroscopic applications. A major challenge in spectroscopic detection is the reduction in signal-to-noise ratio caused by nonspecific molecular adsorption, endogenous background signals (e.g., proteins, lipids), and cellular autofluorescence. Proteins, in particular, present additional complications because of their high molecular weight and structural variability, leading to instability. This issue is further amplified in SERS, where protein detection is influenced not only by their intrinsic polarizability but also by their interactions with nanostructured surfaces and localized electromagnetic fields. For successful clinical adoption, it is essential to develop SERS nanotags that are nontoxic, biodegradable, and efficiently cleared from the body to minimize potential side effects.

Moreover, ethical considerations represent a distinctive and imperative dimension in all medical research, requiring particular attention in Raman spectroscopy applications. The translation of Raman spectroscopy from research to clinical practice involves a rigorous, multistage validation process to ensure safety, efficacy, and regulatory compliance. A typical structured pathway included: (1) proof-of-concept studies establishing technical feasibility using preclinical models from molecules to cells and to animals; (2) preclinical validation on human ex vivo biofluid or tissue; (3) clinical trials (Phases I–III) assessing safety, diagnostic accuracy, and noninferiority to gold-standard methods (e.g., histopathology); (4) regulatory approval (e.g., FDA) demanding standardized protocols and robust evidence of clinical utility; (5) workflow integration necessitating clinician training and cost-effectiveness analyses; (6) postmarket surveillance ensuring real-world reliability, addressing challenges like reproducibility and evolving regulatory frameworks. Compared to in vivo use that is limited by the strict regulations, translation toward SERS point-of-care application in ex vivo biofluid or tissue is more promising. A small sample size is indeed a common limitation in current clinical Raman spectroscopy studies. ,, Therefore, the results of single-center, small-sample clinical studies should be interpreted with caution. For a Raman spectroscopic technique to be translated into routine clinical practice, multicenter and large-sample validation is essential.

AI Implementation Challenges

For Raman spectroscopy’s application in oncology, AI has demonstrated promising potential for transformative integration. However, AI technologies are not without limitations, presenting several non-negligible challenges that warrant careful consideration. A critical barrier to implementing machine learning-enhanced Raman in biomedicine is the scarcity of comprehensive, high-quality data sets required for developing reliable predictive models. Current limitations (1) insufficient sample sizes, particularly for rare conditions or early-stage pathologies; (2) inconsistencies arising from batch effects, instrumental variability, and protocol differences across institutions; and (3) the resultant “localized” machine learning models that exhibit limited generalizability beyond their original experimental setups. ,, The convergence of these limitationsincluding data set biases, methodological inconsistencies, and insufficient prospective validationcompromises model generalizability in clinical settings and heightens risks of overfitting. To ensure the development of clinically reliable AI models, stringent quality control measures must be systematically implemented across all critical phasesincluding but not limited to standardized data acquisition protocols, rigorous model optimization procedures, and comprehensive validation frameworks employing diverse clinical data sets.

Furthermore, ethical imperatives demand solutions for critical AI-related issues. AI utilizing protected health information or genomic data sets must address critical vulnerabilities in data anonymization and storage protocols to prevent potential breaches of patient confidentiality. And the inherent opacity of complex neural network architecturesso-called “black-box” natureposes substantial challenges for clinical adoption. This “explainability gap” obscures the decision-making processes underlying diagnostic predictions, potentially compromising informed consent and eroding physician confidence in AI-assisted outcomes. A promising strategic direction involves creating curated, multicenter SERS spectral databases that standardize heterogeneous data sets, thereby enhancing both the reproducibility of machine learning models and their compliance with healthcare ethics frameworks. , Legal constraints represent another critical challenge that cannot be overlooked in current AI applications for medicine. In August 2024, the EU Artificial Intelligence Act (AI Act)the world’s first comprehensive regulatory framework of AI for healthcarecame into effect. This regulation establishes essential quality and safety standards for AI applications in healthcare, with particular emphasis on data governance frameworks for AI-enabled medical devices. Nevertheless, further collaborate actions like clarification of core definitions to minimize regulatory ambiguity and specification of individual rights for patients are recommended.

Interdisciplinary Gap

The historical development of Raman spectroscopy has fundamentally relied on the convergence and synergy of optical physics, materials science, and other fundamental disciplines. While computer science and AI remain relatively nascent fields, they have already emerged as transformative engines for Raman-based technological innovation. Conversely, the integration of Raman spectroscopy has reciprocally advanced applied sciences including clinical medicine, nutritional science, and environmental studies. The bridging role of Raman spectroscopy across these diverse disciplines is comprehensively demonstrated in Figure of our study.

However, current training systems for researchers reveal critical knowledge fundamental scientists often lack awareness of potential clinical applications for their technical breakthroughs, while medical professionalstrained within specialized and relatively isolated systemsfrequently find “Raman spectroscopy” an unfamiliar concept, remaining unaware of its potential solutions to pressing oncological challenges. To address these disconnects, we propose a dual-pathway strategy: (1) enhancing continuing medical education to expose clinicians to cutting-edge fundamental scientific advances, thereby equipping them to identify novel solutions for clinical problems, and (2) establishing demand-driven research initiatives that assemble interdisciplinary teams working toward clearly defined translational goals. This approach would synergize the complementary expertise of diverse specialists while maintaining a focus on tangible clinical outcomes.

Conclusions

Over the past quarter-century, Raman spectroscopy has evolved from a fundamental analytical technique to a transformative tool in clinical oncology, offering unprecedented opportunities for cancer diagnosis, surgical guidance, and therapeutic monitoring. Our comprehensive bibliometric analysis underscores the remarkable growth of this field, characterized by a 20.73% increase in annual publications and a paradigm shift toward translational applications. The exponential rise in research output, particularly after 2010, reflects the convergence of technological advancementsnotably in SERS, nanotechnology, and AIwith pressing clinical needs in oncology. The geographical and institutional landscape reveals a dynamic interplay of global contributions, with China, the USA, and India leading in productivity. The dominance of SERS and nanoparticle-based strategies in early research phases (2000–2015) has gradually shifted to AI-driven approaches, as evidenced by burst keywords. These methodologies have addressed challenges in spectral interpretation, achieving >95% diagnostic accuracy in some studies and enabling real-time, intraoperative decision-making.

Pioneering work in tumor margin assessment (e.g., brain and breast cancer) demonstrated Raman spectroscopy’s clinical utility, while liquid biopsy applications expanded its minimally invasive potential. Raman spectroscopy has enhanced tumor diagnostic accuracy across multiple biological levels, including metabolic compounds, nucleic acids, proteins, cells, and tissues. Current research hotspots focus on exosomes and intraoperative navigation, where this technology demonstrates transformative potential for clinical oncology practice. The integration of photothermal therapies with Raman-guided precision exemplifies the shift toward theranostic platforms. Raman spectroscopy applications have expanded from clinical oncology to include drug concentration monitoring, environmental contaminant detection, and food safety analysis. The continuous advancement of 3D printing, microfluidic chip technology, and wearable devices enables high-throughput, point-of-care, long-term, and more convenient health monitoring.

Despite these advances, critical challenges remain. Standardization of protocols, reproducibility across diverse cancer types, and cost-effective instrumentation are urgent priorities for the widespread clinical adoption. Furthermore, the ethical and regulatory implications of AI-driven diagnostics warrant a rigorous scrutiny. Future research should focus on large-scale clinical trials to validate Raman-based tools against gold-standard methods as well as on developing portable, user-friendly systems for low-resource settings.

The 50th anniversary of SERS in 2024 marks an opportune moment to reflect on Raman spectroscopy’s journey and envision its future trajectory. By leveraging emerging technologies, the next decade could realize the full potential of this platform for precision oncology. Our analysis not only maps the past and present of Raman spectroscopy in cancer research but also provides a roadmap for scientist, clinicians, and engineers to collaboratively address the unmet needs in cancer care. Its continued evolution will depend on sustained interdisciplinary collaboration, translational investment, and a commitment to equitable healthcare solutions.

Supplementary Material