Authors: Laura Pérez-Carasol, Saul Martinez-Horta, Andrea Horta-Barba, Helena Bejr-Kasem, Juan Marín-Lahoz, Jesús Perez-Perez, Ignacio Aracil-Bolaños, Javier Pagonabarraga, Jaime Kulisevsky
Categories: Article, Diseases, Neurology, Neuroscience
Source: NPJ Parkinson's Disease
Authors: Laura Pérez-Carasol, Saul Martinez-Horta, Andrea Horta-Barba, Helena Bejr-Kasem, Juan Marín-Lahoz, Jesús Perez-Perez, Ignacio Aracil-Bolaños, Javier Pagonabarraga, Jaime Kulisevsky
Minor hallucinations are frequent and clinically relevant in Parkinson’s disease (PD), often preceding cognitive decline and more complex psychotic symptoms. These subtle perceptual anomalies are thought to result from an imbalance between degraded sensory input and dysregulated higher-order cognitive processes. Because visual categorization relies on the integration of perceptual, semantic, and evaluative operations, it provides a powerful framework to investigate the neural mechanisms underlying hallucination vulnerability. Ninety-three non-demented PD patients (mean age = 68.4 years, 41% female) performed a visual categorization task with faces, objects, and face-like stimuli during 19-channel electroencephalography. Patients were classified by hallucination (present/absent) and cognitive status (normal/MCI), yielding four subgroups. Hallucinating patients showed reduced accuracy for faces and objects despite preserved reaction times, with the greatest impairment in those with both hallucinations and cognitive deficits. Event-related potentials revealed reduced N170, enhanced N300, and diminished P600 amplitudes in hallucinating patients, particularly with MCI. Source estimation indicated reduced occipito-temporal activation (N170), premature engagement of default mode and hippocampal regions (N300), and impaired posterior control signals (P600). These findings delineate a progressive disruption of visual-semantic integration in PD hallucinations, reflecting the convergence of impaired sensory encoding, semantic overactivation, and weakened top-down cognitive control, a mechanistic signature of hallucination vulnerability in PD.
Parkinson’s disease (PD) is primarily recognized as a movement disorder characterized by motor symptoms such as rigidity, bradykinesia, and tremor^1^. However, it is well established that PD also encompasses a broad spectrum of non-motor symptoms^2^, with cognitive impairment and minor visual hallucinations being among the most prevalent, and even emerging in the early stages of the disease^3–7^.
The phenomenology of minor hallucinations in PD is defined by subtle, often transient perceptual disturbances that typically arise without apparent external stimuli. These include presence hallucinations like the sensation that someone is nearby when no one is present, passage hallucinations, consisting of brief, fleeting images of shadows or figures in the peripheral vision, and visual illusions, involving the misinterpretation of actual objects^3,4^.
Patients generally retain insight, recognizing these experiences as unreal, and they are usually non-threatening and non-distressing^3^. However, as the disease progresses, insight can gradually deteriorate, particularly as hallucinations become more complex or occur in the context of cognitive impairment^3,4^. The mechanisms supporting or undermining this metacognitive awareness remain unclear, likely involving higher-order monitoring systems that evaluate the coherence between sensory evidence and internal perceptual predictions^8–10^. Minor hallucinations are frequently underreported, either because patients consider them insignificant or due to fear of stigma^11^.
Minor hallucinations are highly prevalent across the clinical spectrum of PD, with rates varying depending on disease stage and cognitive status^3,4,12^. In the prodromal phase, prevalence of minor hallucinations estimates up to 33.3%^7^. In patients with formal clinical diagnosis of PD, the prevalence increases significantly in coexistence with mild cognitive impairment (PD-MCI), affecting 30–50%, and is even higher in those with PD dementia (PDD), where rates may exceed 70%^3,4,7,12^. Moreover, longitudinal studies have consistently shown that minor hallucinations in non-demented PD patients are associated with an increased risk of developing dementia over time^13–15^.
From both a neuroanatomical and functional perspective, minor hallucinations in PD have been linked to structural alterations in posterior cortical regions and to disrupted activity within and between key brain networks, including the default mode network (DMN), the salience network, and multimodal association cortices^4,12,13^. Among the mechanisms proposed, one prominent hypothesis suggests that dysregulated network dynamics may lead to an overreliance on internally generated representations, particularly when bottom-up sensory input is weak or ambiguous^4,16,17^. In this context, the DMN may become inappropriately engaged during the early stages of visual processing, substituting inference-based predictions for accurate sensory encoding^18^. If this premature engagement is not properly regulated by top-down control systems, internally driven interpretations may persist unchecked, thereby facilitating the emergence of hallucinatory experiences^12,18,19^.
Visual categorization is a complex, hierarchical process that enables the brain to organize sensory inputs into meaningful categories based on shared perceptual or conceptual features^20^. Categorization relies on the coordinated engagement of distributed neural systems as well as on temporally structured neurophysiological dynamics that support the progressive transformation of raw sensory input into abstract representations^21–24^.
The temporal unfolding of visual categorization and recognition occurs within approximately 600 milliseconds and can be reliably indexed by event-related potentials (ERPs)^20,24–26^. The earliest stage involves rapid sensory encoding within primary visual and adjacent extrastriate cortices, reflected by the P100 component, a positive occipital deflection emerging around 100 ms post-stimulus, indexing feedforward visual registration^27^. As information progresses along the ventral visual stream, occipito-temporal structures such as the lateral occipital cortex and fusiform gyrus support structural encoding and early categorical discrimination. This stage is captured by the N170 component, a negative deflection between 150 and 240 ms over temporo-occipital sites^28–30^. Although traditionally associated with face processing, the N170 is also elicited by other visually complex stimuli, indicating that it reflects sensitivity to perceptual complexity more than strict category specificity^31–33^. At this stage, the visual system initiates preliminary categorization processing based on structural features, even before full semantic identification^30,34^. Beyond early perceptual stages, higher-order integration engages fronto-temporal and temporo-parietal networks involved in semantic memory access, visual-conceptual matching, and meaning-based categorization^35–37^. These processes are indexed by the N300 component, a fronto-central negative deflection around 300 ms post-stimulus^35,38^. The N300 is thought to reflect the attempted alignment of perceptual input with stored semantic representations, particularly when stimuli are atypical, ambiguous, or contextually incongruent^39^. For example, congruent and familiar stimuli tend to elicit reduced N300 amplitudes, while ambiguous or less prototypical stimuli evoke stronger responses, reflecting increased semantic effort^39–41^. Thus, the N300 represents a transitional stage between perceptual encoding (N170) and full semantic integration, and may be disproportionately engaged in individuals with impaired early visual processing, who rely more heavily on top-down interpretative strategies^42^. In the later stages of visual categorization, associative regions including the medial temporal lobe, precuneus, and prefrontal cortex support reappraisal and ambiguity resolution by integrating contextual and mnemonic information to construct coherent perceptual interpretations^38,43,44^. These functions are indexed by the P600, a late parietal-temporal positive deflection occurring around 600–800 ms post-stimulus, associated with reality monitoring, perceptual revision, and top-down cognitive control^38,43–45^. The P600 is typically enhanced when initial perceptual hypotheses must be revised, particularly in response to stimuli that visually resemble one category but belong to another^46^.
Building on current theoretical models of minor hallucinations in PD, we hypothesize that patients who experience these phenomena will exhibit selective disruptions at distinct stages of the visual categorization process, and that the high temporal resolution of ERPs will allow us to detect altered dynamics in visual-semantic integration. Furthermore, we anticipate that cognitive impairment will modulate the nature and severity of these disruptions, potentially amplifying deficits in both early perceptual and later supervisory processes. To do so, we employed a visual categorization task previously developed by our group and used in earlier studies. In this paradigm, participants are presented with visual stimuli belonging to three categories -objects, faces, and face-like objects- and are asked to classify them as quickly and accurately as possible. In our previous study using this task^47^, healthy control participants performed with high accuracy and consistent reaction times across conditions, confirming the feasibility and reliability of the paradigm for assessing rapid visual categorization processes. Importantly, in that earlier work, we focused specifically on the N170 component, reflecting early perceptual stages of face and object processing. In the current study, however, we aimed to extend this approach by characterizing the entire sequence of neurophysiological events associated with visual categorization. This broader perspective aligns our work with other studies employing similar paradigms to investigate how perceptual, semantic, and higher-order cognitive mechanisms unfold during object. By situating our paradigm in this framework, we formulated directional hypotheses that disruptions in early perceptual and later supervisory stages would differentially relate to the occurrence and severity of minor hallucinations in PD. By addressing these hypotheses, our goal is to advance the understanding of the neurocognitive architecture underlying minor hallucinations in PD, and to clarify how these experiences interact with broader cognitive dysfunction across the disease spectrum.
The final sample consisted of 93 participants, of whom 55 were classified as PD-NH and 38 as PD-MH. Among the PD-MH group, 34.2% (n = 14) met criteria for PD-MCI, compared to 21.8% (n = 12) in the PD-NH group. This difference in PD-MCI distribution between groups was not statistically significant (χ² = 1.18, p = 0.277).
Demographic and clinical characteristics of the two groups (PD-NH, and PD-MH) are summarized in Table 1. No significant differences were observed across groups for age, sex, or education. Disease duration was significantly longer in PD-MH (U = 1423.5, p = 0.005), whereas UPDRS-III scores, LEDD, PD-CRS total, and subscale scores showed no group differences.Table 1Clinical and sociodemographic data of the samplePD-NHPD-MHp valueAge (years)66.31 (8.44)68.70 (6.98)0.141Sex (% male / % female)58.2/41.871.1/28.90.295Disease duration (years)4.75 (2.84)6.64 (3.30)0.005Education (years)12.31 (4.96)13.11 (4.49)0.423UPDRS III^a^25.82 (8.01)24.79 (7.85)0.539LEDD^b^149.82 (103.14)161.24 (126.49)0.647PD-CRS Total score^c^93.98 (15.93)89.42 (15.00)0.163PD-CRS Frontal-subcortical65.87 (15.03)60.61 (14.30)0.091PD-CRS Posterior-cortical28.11 (2.02)28.82 (1.59)0.063^a^Unified Parkindon’s disease rating scale—part III.^b^L-dopa equivalent daily dose. ^c^Parkinson’s disease—cognitive rating scale total score.
Among PD-MH patients, presence of hallucinations was the most frequent, followed by passage hallucinations and pareidolic visual misperceptions, while auditory, olfactory, and moderately complex visual hallucinations were rare, and tactile hallucinations were absent.
Focusing on diagnostic subgroups, 43 participants were classified as PD-NH/MCI–, 12 as PD-NH/MCI+, 25 as PD-MH/MCI–, and 14 as PD-MH/MCI+. Although the subgroup comparisons reflect a 2 × 2 structure (hallucination status × cognitive status), the analyses were conducted using one-way ANOVAs to account for the small sample size of some subgroups and to maintain clinical interpretability. This approach allowed examination of the additive and combined effects of hallucinations and cognitive impairment while minimizing violations of statistical assumptions. One-way ANOVAs revealed significant group effects for age [F(3,89) = 6.51, p = 0.0005], disease duration [F(3,89) = 3.20, p = 0.027], education [F(3,89) = 3.25, p = 0.026], UPDRS-III [F(3,89) = 2.76, p = 0.047], and cognitive performance on the PD-CRS total [F(3,89) = 38.58, p < 0.0001], frontal–subcortical [F(3,89) = 37.05, p < 0.0001], and posterior–cortical subscores [F(3,89) = 7.52, p < 0.001]. Post hoc analyses indicated that PD-NH/MCI+ patients were younger and had longer disease duration than other groups and fewer years of education compared to PD-NH/MCI– and PD-MH/MCI–. Motor severity was greater in PD-MH/MCI– than in PD-MH/MCI+, and cognitive scores were consistently lower in both MCI+ groups relative to MCI– groups. Detailed descriptive data and statistical comparisons are provided in Table 2. Detailed analyses of the co-occurrence of multiple hallucination types and the frequency distribution of each phenomenon across PD-MCI subgroups are provided in Supplementary Materials S2–S3.Table 2Clinical and sociodemographic data of the different clinical subgroupsGroups–mean (SD)Post-hoc comparisonsVariablePD-NH/MCI-PD-MH/MCI-PD-NH/MCI+PD-MH/MCI+p valuePD-NH/MCI- vs PD-MH/MCI-PD-NH/MCI- vs PD/NH/MCI+PD-NH/MCI- vs PD-MH/MCI+PD-MH/MCI- vs PD-NH/MCI+PD-MH/MCI- vs PD-MH/MCI+PD-NH/MCI+ vs PD-MH/MCI+Age (years)64.64 (8.44)66.48 (7.35)72.32 (5.26)72.95 (3.50)<0.0010.349<0.001<0.0010.009<0.0010.730Disease duration (years)4.91 (2.93)6.40 (3.05)4.20 (2.52)7.11 (3.83)0.0270.0530.4160.0740.0280.5730.034Education (years)13.23 (4.57)13.68 (4.82)9.00 (5.10)12.00 (3.72)0.0250.7080.0190.3310.0140.2430.110UPDRS-III25.09 (7.60)22.56 (7.85)28.42 (9.23)29.08 (6.03)0.0470.2000.2700.0610.0730.0070.836LEDD158.28 (108.06)176.04 (138.87)119.50 (79.72)132.77 (97.11)0.4630.9980.8250.8710.7480.8220.998PD-CRS Total score100.21 (11.65)97.36 (12.00)71.67 (5.74)74.15 (4.67)<0.0010.344<0.001<0.001<0.001<0.0010.250PD-CRS Frontal-subcortica71.63 (11.30)68.16 (11.46)45.25 (5.29)46.08 (4.54)<0.0010.232<0.001<0.001<0.001<0.0010.680PD-CRS Posterior-cortical28.58 (1.59)29.20 (1.22)26.42 (2.54)28.08 (1.98)<0.0010.0770.0140.4120.0030.0780.084
A 2 (group: PD-MH, PD-NH) × 3 (stimulus faces, face-like objects, objects) repeated-measures ANOVA revealed a robust main effect of stimulus type on accuracy rates [F(2,184) = 60.92, p < 0.0001]. Table 3 shows all the data related to the behavioral parameters obtained. Post hoc paired comparisons confirmed that participants were significantly more accurate when categorising faces compared with face-like objects [t(92) = −8.99, p < 0.0001] and objects [t(92) = −4.02, p = 0.0001]. Accuracy was also higher for objects than for face-like objects [t(92) = 6.94, p < 0.0001], indicating that face-like objects elicited the greatest ambiguity and error susceptibility. Analysis of RTs revealed a strong main effect of stimulus type [F(2,184) = 262.57, p < 0.0001]. Participants were significantly slower to faces compared with objects [t(92) = 19.99, p < 0.0001] and face-like objects [t(92) = 19.02, p < 0.0001]. No significant RT differences were found between objects and face-like objects. When restricting analysis to correct trials, a similar pattern was observed [F(2,184) = 165.43, p < 0.0001]. Responses to faces were significantly slower than to either objects [t(92) = 19.49, p < 0.0001] or face-like objects [t(92) = 13.90, p < 0.0001], while no significant difference emerged between object and face-like object trials. Stimulus type also significantly affected RTs for incorrect responses [F(2,166) = 59.30, p < 0.0001]. Participants were significantly faster when making errors to face-like objects compared with objects [t(83) = −6.60, p < 0.0001], but slower when responding incorrectly to faces than to either face-like objects [t(83) = 10.51, p < 0.0001] or objects [t(83) = 4.41, p < 0.0001].Table 3Behavioral performance on the visual categorization taskMean (SD)PD-NHPD-MHp valueProportion of correct responsesFace0.97 (0.03)0.91 (0.16)0.022Object0.93 (0.05)0.88 (0.13)0.015Face-like object0.66 (0.32)0.65 (0.31)0.888Reaction times (ms)Face0.50 (0.08)0.52 (0.08)0.189Correct0.50 (0.08)0.53 (0.08)0.161Incorrect0.47 (0.13)0.46 (0.10)0.665Object0.60 (0.09)0.61 (0.09)0.403Correct0.60 (0.09)0.62 (0.10)0.289Incorrect0.53 (0.12)0.54 (0.10)0.648Face-like object0.60 (0.09)0.61 (0.11)0.443Correct0.59 (0.10)0.61 (0.12)0.404Incorrect0.62 (0.14)0.63 (0.15)0.859
Focusing on the comparison between patients with and without minor hallucinations, univariate ANOVAs revealed significant group differences in accuracy for face stimuli [F(1,91) = 5.36, p = 0.023] and object stimuli [F(1,91) = 6.08, p = 0.016], indicating reduced accuracy in PD-MH patients. No significant differences were observed for accuracy on face-like objects stimuli or for overall reaction times to any stimulus type. Similarly, reaction times for correct and incorrect responses across all stimulus types did not differ significantly between groups. Figure 1 depicts the behavioral parameters between groups.Fig. 1Behavioral performance.A Behavioral performance on the visual categorization task in PD-NH and PD-MH. Performance across conditions and groups, showing accuracy and reaction times. Accuracy (left) is expressed as the percentage of correct responses. Reaction times (RTs), collapsed across correct and incorrect trials, are represented in milliseconds. Asterisks indicate statistically significant differences (p < 0.05). B Behavioral performance on the visual categorization task in the clinical subgroups. Performance across conditions and groups, shown in terms of accuracy and reaction times. Accuracy (left) is expressed as the percentage of correct responses. Reaction times (RTs), collapsed across correct and incorrect trials, are represented in milliseconds. Asterisks indicate statistically significant differences (p < 0.05).
Repeated measures ANOVA revealed significant main effects of stimulus type [F(2,178) = 65.84, p < 0.0001] and clinical group [F(3,92) = 14.41, p < 0.0001], as well as a significant interaction between the two factors [F(6,178) = 3.48, p = 0.002]. Post hoc comparisons revealed that PD-MH/MCI+ consistently underperformed, particularly for face and face-like stimuli, showing significantly lower accuracy compared to all other subgroups. Notably, PD-NH/MCI+ also performed worse than PD-NH/MCI− across several stimulus categories, suggesting an additive effect of cognitive impairment. For object stimuli, both PD-MH/MCI+ and PD-NH/MCI+ exhibited reduced performance relative to PD-NH/MCI−.
Reaction time (RT) analyses mirrored these findings. Significant main effects of both stimulus type [F(2,178) = 262.10, p < 0.0001] and clinical group [F(3,92) = 74.64, p < 0.0001] were observed, although the interaction was not significant. As detailed in Table 4, PD-MH/MCI+ showed the longest response times across all conditions, followed by PD-NH/MCI+. In contrast, PD-MH/MCI– and PD-NH/MCI– performed comparably, suggesting that MCI status is a stronger determinant of RT impairment than hallucination status alone.Table 4Behavioral performance in the clinical subgroupsGroups–mean (SD)Post-hoc comparisonsPD-NH/MCI-PD-MH/MCI-PD-NH/MCI+PD-MH/MCI+P valuePD-NH/MCI- vs PD-MH/MCI-PD-NH/MCI- vs PD/NH/MCI+PD-NH/MCI- vs PD−MH/MCI+PD-MH/MCI- vs PD-NH/MCI+PD-MH/MCI- vs PD-MH/MCI+PD-NH/MCI+ vs PD-MH/MCI+Proportion of correct responsesFace0.98 (0.02)0.94 (0.12)0.93 (0.03)0.86 (0.21)0.0080.158;0.095 ^pc^0.778;1.000 ^pc^0.288;1.000 ^pc^0.001;0.006 ^pc^0.075;0.454 ^pc^<0.001; 0.004 ^pc^Object0.93 (0.06)0.89 (0.07)0.92 (0.04)0.86 (0.21)0.0780.481;1.000 ^pc^0.115;0.691 ^pc^0.282;1.000 ^pc^0.047;0.284 ^pc^0.015;0.091 ^pc^0.865;1.000 ^pc^Face-like object0.72 (0.29)0.74 (0.28)0.43 (0.35)0.37 (0.32)0.0020.009;0.056 ^pc^0.006;0.036 ^pc^0.789;1.000 ^pc^0.009;0.058 ^pc^0.771;1.000 ^pc^0.005;0.0325 ^pc^Reaction times (ms)Face0.48 (0.06)0.49 (0.06)0.56 (0.10)0.58 (0.08)<0.001<0.001;0.004 ^pc^0.017;0.103 ^pc^0.585;1.000 ^pc^<0.001; <0.001 ^pc^0.582;1.000 ^pc^0.002;0.011 ^pc^Correct0.49 (0.06)0.49 (0.06)0.56 (0.10)0.59 (0.09)<0.001<0.001;0.004 ^pc^0.015;0.095 ^pc^0.527;1.000 ^pc^<0.001; <0.001 ^pc^0.556;1.000 ^pc^0.001;0.008 ^pc^Incorrect0.45 (0.11)0.45 (0.11)0.53 (0.17)0.47 (0.08)0.2070.7388;1.000 ^pc^0.114;0.685 ^pc^0.250;1.000 ^pc^0.621;1.00 ^pc^0.863;1.000 ^pc^0.060;0.364 ^pc^Object0.58 (0.09)0.58 (0.07)0.66 (0.10)0.67 (0.10)<0.0010.002;0.017 ^pc^0.008;0.049 ^pc^0.776;1.000 ^pc^0.001;0.011 ^pc^0.857;1.000 ^pc^0.006;0.040 ^pc^Correct0.58 (0.08)0.59 (0.07)0.67 (0.10)0.68 (0.11)<0.0010.005;0.031 ^pc^0.012;0.077 ^pc^0.726;1.000 ^pc^0.001;0.008 ^pc^0.659;1.000 ^pc^0.005;0.032 ^pc^Incorrect0.52 (0.12)0.53 (0.09)0.55 (0.14)0.55 (0.11)0.6810.474;1.000 ^pc^0.522;1.000 ^pc^0.985;1.000 ^pc^0.348;1.00 ^pc^0.708;1.000 ^pc^0.382;1.000 ^pc^Face-like object0.58 (0.09)0.57 (0.07)0.66 (0.10)0.69 (0.13)<0.001<0.001;0.004 ^pc^0.004;0.025 ^pc^0.498;0.1 ^pc^<0.001;0.004 ^pc^0.745;1.000 ^pc^0.01;0.060 ^pc^Correct0.58 (0.09)0.57 (0.08)0.63 (0.11)0.69 (0.14)<0.001<0.001;0.003 ^pc^0.030;0.180 ^pc^0.239;1.000 ^pc^<0.001;0.004 ^pc^0.540;1.000 ^pc^0.0761;0.456 ^pc^Incorrect0.60 (0.12)0.60 (0.15)0.71 (0.18)0.69 (0.14)0.0340.081;0.485 ^pc^0.063;0.378 ^pc^0.789;1.000 ^pc^0.0296;0.177 ^pc^0.939;1.000 ^pc^0.019;0.116 ^pc^pc Corrected p value.
These group differences were also evident when analyses were restricted to correct responses [F(3,92) = 46.63, p < 0.0001], with PD-MH/MCI+ and PD-NH/MCI+ both showing prolonged RTs. No significant differences emerged between PD-MH/MCI– and PD-NH/MCI–. A similar pattern was observed for incorrect responses [F(3,92) = 5.62, p = 0.001], with the PD-MH/MCI+ group again showing the slowest responses. All post hoc comparisons were corrected using Tukey’s HSD. Full statistics for pairwise contrasts are reported in Table 4.
In a subsequent analysis, we examined whether specific hallucinatory subtypes were associated with performance on the visual categorization task using univariate linear regressions. Each model included a single hallucination type as a predictor of accuracy or RT. Across all analyses, none of the hallucination subtypes showed statistically significant associations with either accuracy or reaction time for any stimulus condition. Given the small number of participants within certain hallucinatory subtypes, these analyses were likely underpowered and should therefore be interpreted with caution.
The number of artifact-free EEG epochs (mean ± SD) included in the averages did not differ significantly between groups or conditions. For the Face controls = 139.33 ± 0.96, PD = 139.69 ± 1.94, PD–NH = 139.87 ± 2.34, PD–MH = 139.42 ± 1.11, PD-NH/MCI– = 139.91 ± 1.06, PD-NH/MCI + = 139.75 ± 4.75, PD-MH/MCI– = 139.48 ± 1.16, and PD-MH/MCI + = 139.31 ± 1.03. For the Object controls = 139.33 ± 0.92, PD = 139.65 ± 1.88, PD–NH = 139.80 ± 2.28, PD–MH = 139.42 ± 1.06, PD-NH/MCI– = 139.88 ± 0.98, PD-NH/MCI + = 139.50 ± 4.66, PD-MH/MCI– = 139.52 ± 1.16, and PD-MH/MCI + = 139.23 ± 0.83. For the Face-like objects controls = 139.00 ± 1.93, PD = 139.75 ± 1.75, PD–NH = 139.89 ± 2.13, PD–MH = 139.55 ± 0.95, PD-NH/MCI– = 139.88 ± 1.12, PD-NH/MCI + = 139.92 ± 4.19, PD-MH/MCI– = 139.64 ± 0.91, and PD-MH/MCI + = 139.38 ± 1.04. No significant differences were observed between groups or conditions (all p > 0.8), and no participants or channels were excluded due to insufficient valid trials, confirming that trial retention was uniformly high and balanced across the entire sample.
Visual inspection of ERP waveforms revealed a consistent sequence of components reflecting distinct stages of visual processing (Fig. 2). The earliest was the P100, a positive deflection around 100 ms over occipital electrodes. This was followed by the N170, a negative deflection at 150–240 ms over temporo-occipital regions. The N300 was found as a fronto-central negativity at 250–350 ms. Finally, the P600, a late positive component at 450–750 ms, appeared over posterior sites.Fig. 2Main ERPs components.ERP components elicited at successive time points during visual categorization. Topographical maps illustrate the scalp distribution of each component at the corresponding latency.
To evaluate the modulation of ERP component amplitudes by stimulus type, repeated-measures ANOVAs were conducted separately for each component. Post hoc pairwise comparisons were performed using paired t-tests with Bonferroni correction to control for multiple comparisons.
P100 amplitude showed a significant main effect of stimulus type [F(2,184) = 34.72, p < 0.0001]. Post hoc tests revealed that amplitudes were significantly greater for face stimuli compared to both objects [t(92) = 7.05, p < 0.001] and face-like objects [t(92) = 6.89, p < 0.001], with no significant difference between objects and face-like objects [t(92) = -0.71, p = 0.48]. N170 amplitude was strongly modulated by stimulus type [F(2,184) = 142.26, p < 0.0001]. Faces elicited significantly larger N170 amplitudes than both objects [t(92) = 11.95, p < 0.001] and face-like objects [t(92) = 13.32, p < 0.001], while no difference was found between objects and face-like objects [t(92) = 0.40, p = 0.69]. N300 amplitude also showed a significant effect of stimulus type [F(2,184) = 13.11, p < 0.0001]. Post hoc comparisons indicated that face-like objects elicited larger N300 amplitudes than objects [t(92) = −6.33, p < 0.001] and faces [t(92) = −6.44, p < 0.001], whereas the difference between faces and objects was not significant [t(92) = −1.44, p = 0.15]. P600 amplitude differed by stimulus type [F(2,184) = 9.74, p = 0.0001]. Faces evoked larger P600 amplitudes compared to objects [t(92) = −2.23, p = 0.028], and face-like objects elicited larger amplitudes than both faces [t(92) = −3.90, p < 0.001] and objects [t(92) = −2.69, p = 0.009].
We examined correlations between ERP amplitudes and age, disease duration, UPDRS-III scores, and LEDD to assess potential clinical and demographic influences. Although disease duration differed statistically between groups, the absolute difference was modest, and correlation analyses revealed no significant or consistent associations between disease duration and ERP amplitudes or behavioral outcomes. Therefore, in line with standard recommendations for ANCOVA and repeated-measures ANOVA designs, these variables were not included as covariates in subsequent analyses.
The number of EEG epochs included in the averages did not differ between groups (Faces: PD-MH = 139.31 ± 0.92; PD-NH = 139.32 ± 3.20. Objects: PD-MH = 139.17 ± 0.65; PD-NH = 139.82 ± 3.10; Face-like PD-MH = 139.21 ± 0.86; PD-NH = 139.52 ± 3.60). To assess the impact of minor hallucinations on ERPs responses, we conducted a factorial linear model including ERP component, stimulus type, and group. This model revealed a significant main effect of group [F(3,198) = 3.90, p = 0.01], indicating that, as seen in Fig. 3, overall ERP amplitudes differed between PD-MH and PD-NH participants. In contrast, the stimulus type × group interaction was not significant. Post-hoc comparisons focusing on each specific ERP component showed no significant group differences in P100 amplitudes. However, for the remaining components, divergent group effects were observed. Specifically, for the N170, PD-MH patients exhibited significantly smaller amplitudes across all stimulus types [t(66) = 2.91 p = 0.005]. Conversely, the N300 component was significantly increased across all stimulus conditions in the PD-MH group [t(66) = 3.50, p < 0.001]. Finally, the P600 component was significantly reduced across all stimulus conditions in PD-MH participants [t(66) = 3.50, p < 0.001].Fig. 3ERPs and topographical maps comparisons between PD-NH and PD-MH.Morphology and topographical distribution of ERP components in PD-NH and PD-MH across stimulus categories. A N170 at T6 with corresponding scalp topographies. B N300 at Fz with associated topographical maps. C P600 at Pz with corresponding scalp distribution.
Given that previous ERP analyses revealed no significant group-by-stimulus type interactions, we focused on the neurophysiological dynamics underlying the time course of visual categorization, irrespective of stimulus category. To this end, ERP responses were collapsed across all stimulus types into a single composite waveform for each component, allowing characterization of general categorization dynamics without category-specific confounds.
No significant differences were observed in the P100 component across groups. In contrast, as shown in Fig. 4, the N170 component exhibited a progressive, stepwise reduction in amplitude across groups [F(3,89) = 2.73, p = 0.048]. Post hoc comparisons revealed a stepwise pattern of significant the PD-NH/MCI- group exhibited the highest N170 amplitude, significantly greater than that of the PD-MH/MCI- group [t(89) = 2.10, p = 0.038]; the PD-NH/MCI+ group showed a further decrease compared to PD-MH/MCI- [t(89) = 1.98, p = 0.051]; and the PD-MH/MCI+ group exhibited a pronounced reduction relative to PD-NH/MCI+ [t(89) = 2.65, p = 0.010].Fig. 4ERPs, topographical maps and source localization comparisons between clinical subgroups.Morphology of ERP components across groups. A N170 shows a clear gradual reduction across groups. N300 is markedly increased in the PD-MH/MCI+ group. P600 is reduced in PD-NH/MCI+ and PD-MH/MCI+. B Butterfly plots of all components. sLORETA-based source reconstructions are shown for each component and group, with labels indicating regions showing apparent between-group differences.
For the N300 component, a significant main effect of group was found [F(3,89) = 4.56, p = 0.005], with post hoc analyses indicating that this effect was primarily driven by a pronounced increase in N300 amplitude in the PD-MH/MCI+ group relative to the PD-NH/MCI− [t(89) = 3.21, p = 0.002] and PD-MH/MCI- groups [t(89) = 2.85, p = 0.006].
Regarding the P600 component, a main effect of group was also observed [F(3,89) = 3.89, p = 0.012]. Post hoc comparisons showed a comparable reduction of P600 amplitude in both the PD-MH/MCI+ and PD-NH/MCI+ groups relative to the PD-NH/MCI– [t(89) = –2.75 and –2.25, p = 0.008 and 0.028, respectively]. The PD-MH/MCI– group demonstrated a trend toward increased P600 amplitude compared to PD-NH/MCI– [t(89) = 1.92, p = 0.058], which did not reach statistical significance. Additional linear regression models were conducted to explore whether specific hallucinatory subtypes were differentially associated with ERP amplitudes while accounting for cognitive status. The detailed results of these analyses are provided in the Supplementary Materials S4.
Source localization analysis using sLORETA revealed distinct and progressively divergent patterns of source activation across the three ERP components and the four clinical groups. For the N170 component, source activity in the PD-NH/MCI− group was localized bilaterally to occipito-temporal cortices. A progressive reduction in N170-related activation was observed in the PD-NH/MCI+ and PD-MH/MCI− groups, with the PD-MH/MCI+ group exhibiting markedly diminished occipito-temporal activity. Unlike the limited and more spatially restricted activity seen in the other groups, N300 sources in PD-MH/MCI+ extended beyond fronto-medial and parietal cortices, including regions overlapping with DMN, to bilateral hippocampal areas. For the P600 component, robust activation was observed in posterior parietal and occipital cortices in the PD-NH/MCI− group, and this pattern was largely preserved in the PD-MH/MCI− group. In contrast, both PD-NH/MCI+ and PD-MH/MCI+ groups exhibited significant reductions in P600 amplitude.
This study comprehensively examined the behavioral and neurophysiological correlates of minor visual hallucinations with and without cognitive impairment in PD during a visual categorization task. Our main findings illuminate the complex interplay between perceptual processing deficits and high-order cognitive dysfunction, contributing to the pathophysiology of hallucinations in PD.
Across the cohort, performance on the visual categorization task reflected expected patterns consistent with the complexity and ambiguity of the stimuli^47^. Accuracy was highest for faces, lower for objects, and lowest for face-like objects, indicating greater perceptual uncertainty with ambiguous stimuli. Reaction times showed the opposite pattern, being slowest for faces, faster for objects, and fastest for face-like objects. Errors on face-like objects were unusually rapid, suggesting impulsive responses to ambiguous cues, whereas errors on faces were slower, possibly reflecting greater hesitation or effortful but unsuccessful processing.
When comparing patients with and without minor hallucinations, those with hallucinations showed significantly reduced accuracy for both faces and objects, despite no significant differences in reaction times. This dissociation indicates that PD-MH patients exhibit specific disruptions in visual discrimination that cannot be explained by general slowing or decisional impulsivity. Rather, the pattern points to deficits in early or mid-level visual encoding, affecting structurally rich stimuli like faces and objects, which depend heavily on bottom-up sensory analysis^29,35^. Interestingly, despite their ambiguity, performance on face-like objects was preserved in the PD-MH group. This selective sparing suggests that ambiguous stimuli may benefit from overactive top-down perceptual priors, particularly face-detection biases^48,49^. In other words, patients prone to hallucinations may rely more heavily on internal templates or expectations when sensory evidence is weak or ambiguous^50^. While this compensatory reliance may support performance for face-like object stimuli, it appears insufficient to maintain accuracy for more concrete visual categories, where precise structural encoding is required.
Further stratification by cognitive status revealed that patients with both hallucinations and mild cognitive impairment (PD-MH/MCI+) exhibited the most pronounced behavioral deficits, with reduced accuracy across all stimulus types and longer reaction times. Patients with MCI but no hallucinations (PD-NH/MCI+) also showed intermediate impairments, particularly for faces and objects. These findings highlight an additive effect of hallucinations and cognitive decline on visual processing^3–5,13^.
Although more complex or multimodal hallucinations were more frequent among patients with cognitive impairment, we interpret this not as a confounding factor but as a reflection of a clinical and neurophysiological continuum. Increasing hallucination complexity in the context of cognitive decline likely reflects a progressive breakdown of visual-to-semantic and supervisory control processes. Accordingly, the present findings support a dimensional model in which the phenomenological richness of hallucinations parallels the severity of underlying neurocognitive dysfunction.
Visual inspection of the grand-averaged ERPs across all participants revealed the expected sequence of components underlying visual categorization^20,24,29,35^. The earliest component, the P100, appeared as a robust positive deflection around 100 ms at occipital electrodes, consistent with initial sensory encoding. The N170 component was clearly modulated by stimulus category, with faces eliciting the largest amplitudes, consistent with its role in structural encoding and early categorical discrimination^33,51^. Following this, the fronto-central N300 component increased for more ambiguous stimuli, particularly face-like objects, reflecting greater perceptual evaluation demands^24,39^. Finally, the posterior P600 component was modulated by stimulus ambiguity, with face-like objects eliciting larger amplitudes than both faces and objects, and faces evoking intermediate responses, indicating a greater engagement of late-stage re-evaluative processes when early perceptual and semantic cues are insufficient to resolve uncertainty^52^.
Comparing patients with and without minor hallucinations, we observed a distinct pattern of ERPs modulations. The P100 remained unaffected, indicating intact early sensory registration across groups. In contrast, the N170 was significantly attenuated in PD-MH patients across all stimulus categories, indicating a disruption in early visual encoding mechanisms. This reduction likely reflects diminished sensory precision at the stage of structural analysis, possibly contributing to perceptual instability. Conversely, the N300 was increased in the PD-MH group, pointing to compensatory or exaggerated recruitment of semantic-associative systems, likely reflecting attempts to resolve ambiguous stimuli through top-down meaning attribution. Finally, the P600 amplitude was reduced in PD-MH patients, suggesting a disruption in late-stage supervisory processes, including conflict detection and reality monitoring^24,43,44,46^.
When stratifying the sample into the four clinical subgroups, these effects followed a more differentiated trajectory. The N170 showed a stepwise reduction from PD-NH/MCI– to PD-MH/MCI+, reflecting progressively impaired perceptual encoding as hallucinations and cognitive impairment co-occur. Notably, only the PD-MH/MCI+ subgroup exhibited significantly elevated N300 amplitudes, suggesting a compensatory reliance on stored semantic representations when ambiguous sensory input coincides with degraded early visual processing and reduced cognitive control. In contrast, P600 amplitude was reduced in both MCI subgroups, regardless of hallucination status, suggesting that cognitive decline exerts a predominant influence on late-stage evaluative and supervisory functions. Although behavioural accuracy differed between groups as a function of stimulus type, ERP analyses revealed group effects that were not stimulus-specific. This apparent dissociation likely reflects the difference between the temporal precision of neural measures and the integrative nature of behavioural outcomes. ERPs capture discrete stages of perceptual and semantic processing, which may be broadly disrupted in patients with hallucinations, whereas behavioural accuracy reflects the cumulative result of these processes. Accordingly, hallucination-related alterations in neural dynamics may exert global effects across categories, while their behavioural manifestation becomes more evident in stimuli with higher perceptual demands.
The sLORETA source reconstruction revealed the neural architecture underlying ERP alterations across clinical subgroups. Although source reconstructions were based on 19-channel EEG data and are therefore limited in spatial precision, the resulting activation patterns were biologically plausible and consistent with established cortical generators for these ERP components. These findings should be interpreted as exploratory and hypothesis-generating, not as statistical evidence of group-level differences. Patients with both hallucinations and cognitive impairment showed progressively reduced activation in bilateral occipito-temporal cortices, paralleling N170 attenuation. In this group, the N300 also displayed a distinct profile, with early recruitment of medial frontal, parietal (DMN), and hippocampal regions, a pattern absent in other subgroups. By contrast, DMN activation in non-hallucinating, cognitively intact participants emerged only at later processing stages, consistent with its role in post-perceptual integration and monitoring.
The premature engagement of DMN regions in the PD-MH/MCI+ group during an intermediate semantic processing window suggests a breakdown in the normal temporal sequencing of network dynamics^4,12,16^. Coactivation of hippocampal regions further points to a pathological coupling between semantic and mnemonic systems, potentially reflecting overreliance on internally generated associations when sensory evidence is weak or ambiguous^53^. We interpret this as a maladaptive top-down cascade in which ambiguous stimuli trigger meaning-laden inferences no longer constrained by early perceptual precision or late-stage evaluative control. P600-related sources highlight additional disruptions in late-stage cognitive control. In patients without cognitive impairment, especially the PD-NH/MCI− group, robust posterior parietal, medial frontal (DMN), and hippocampal activation indicated a fully engaged reality-monitoring network supporting contextual re-evaluation and resolution of perceptual ambiguity. Notably, this pattern was unique to the PD-NH/MCI− group and absent in all other subgroups, where marked attenuation of P600-related activity signaled impaired high-level supervisory control.
Together, these ERP and source-level findings delineate a triple dissociation in neural (1) weakened early visual encoding, (2) overactive reliance on/or dysfunctional semantic-mnemonic integration, and (3) impaired late-stage supervisory control. This configuration was most pronounced in the PD-MH/MCI+ group and supports a mechanistic model in which hallucinations emerge from the convergence of degraded sensory input, dysfunctional semantic attribution, and insufficient cognitive oversight. While this framework reflects a temporally cascading sequence of processing stages, it is important to emphasize that visual perception operates within a highly interactive and parallel architecture. In particular, top-down signals can influence neural responses at the earliest stages of visual processing, even prior to stimulus onset^54^. From this perspective, the disruption observed in patients with hallucinations may not simply reflect a late-stage failure to re-evaluate ambiguous percepts, but rather a misalignment across processing levels, in which top-down influences are mistimed, miscalibrated, or insufficiently constrained, ultimately distorting the interpretation of sensory signals from the outset. This insight reframes the observed triple dissociation not as a strictly serial failure, but as the breakdown of a coordinated and reciprocal interplay between perceptual encoding, semantic integration, and cognitive control.
Secondary analyses showed that most hallucination subtypes shared a common neurophysiological profile. Across visual subtypes, reduced N170, increased N300, and diminished P600 amplitudes were consistently observed, with only minor variations across sensory modalities. This convergence suggests a shared dysfunctional architecture underlying hallucinatory phenomena in PD, rather than entirely distinct mechanisms.
A critical clinical dimension of hallucinatory phenomena in PD is the degree of insight patients retain into the nature of their experiences. Minor hallucinations are typically accompanied by preserved insight, whereby individuals recognize that the perceptual content is not real^3,4,55^. However, as cognitive decline progresses, particularly in the context of dementia, this insight is often lost^4,55^. The mechanisms supporting or undermining metacognitive awareness remain unclear, but our findings offer clues to how insight loss may emerge. We propose that the preserved P600 signal in patients without MCI reflects intact supervisory evaluation, allowing detection of mismatches between internal content and external reality. This function may sustain insight by enabling attentional redirection or cognitive reappraisal after a perceptual anomaly, or by tagging the experience as self-generated even if it cannot be overridden. Thus, intact late-stage monitoring may provide a neurocognitive scaffold for maintaining insight. The reduced P600 in patients with MCI suggests this mechanism is vulnerable, and its decline may precede clinical loss of insight, positioning the P600 as a potential marker of cognitive control and metacognitive awareness.
Several limitations of the present study should be considered. First, although our task was designed to probe neurocognitive processes associated with perceptual vulnerability, no participants experienced overt hallucinations during task execution. Thus, the findings should not be interpreted as reflecting real-time hallucinatory states, but rather as indirect markers of latent susceptibility, revealed by experimentally engaging neural systems hypothesized to be vulnerable in patients with minor hallucinations. Second, while the overall sample size was sufficient to detect group-level effects, subgroup analyses may have been underpowered to detect subtle or interaction-level differences. Larger, more balanced samples will be needed to confirm and expand upon these findings. Third, the cross-sectional design limits causal inferences regarding the progression from perceptual alterations to hallucination onset and cognitive decline. Longitudinal studies will be essential to characterize trajectories and identify early predictive markers. Finally, although sLORETA allowed for cortical source estimation, the use of a 19-channel EEG system imposes limited spatial resolution. While our results align with known functional neuroanatomy, validation using higher-density EEG, MEG, or multimodal imaging would improve anatomical precision and allow for more detailed analyses of functional connectivity dynamics during perception and ambiguity resolution.
Despite these limitations, the study presents several notable strengths. It benefits from the precise identification of patients with and without minor hallucinations, alongside detailed phenotyping of hallucinatory subtypes and their neurophysiological correlates. By capturing the nuanced interplay between hallucination type, cognitive status, and electrophysiological response, the study provides novel evidence of the distinct mechanisms underlying visual hallucinations in PD and how these mechanisms are progressively altered in the context of cognitive decline.
In sum, our findings delineate a structured cascade of neurophysiological alterations underlying visual hallucinations in PD, characterized by early impairments in visual encoding, increased reliance on semantic mechanisms, and compromised late-stage cognitive control. This sequence becomes progressively more pronounced in the presence of cognitive impairment, supporting a model in which hallucinations emerge from the convergence of bottom-up degradation and top-down dysregulation across distributed cortical networks. By characterizing distinct yet overlapping ERP signatures across patient subgroups and hallucination subtypes, we provide empirical support for a multi-level framework that captures both the vulnerability and variability of hallucinatory experiences in PD, anchoring them in a disrupted trajectory of visual-semantic integration. Beyond addressing the core experimental hypotheses, these findings also offer a conceptual foundation for investigating insight from a neurophysiological perspective, suggesting that specific disruptions in late-stage monitoring may play a key role in the transition from preserved to impaired awareness, a hypothesis that future studies are well-positioned to explore.
We conducted a prospective study of 93 consecutive outpatients regularly attending the Movement Disorders Unit at Hospital de la Santa Creu i Sant Pau who fulfilled the diagnostic criteria for Parkinson’s disease. All participants were under regular clinical follow-up within the unit and were prospectively recruited during routine visits. Each patient, accompanied by a caregiver when appropriate, was interviewed by the treating neurologist (J.P., H.B.K., or J.M.L.) regarding disease onset—defined as the estimated time of appearance of the first cardinal motor symptoms rather than the point of formal diagnosis—and medication history. The same neurological team administered the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) and collected all clinical and sociodemographic data, including disease duration, medication type, and dosage. Global cognition was assessed by trained neuropsychologists (S.M.H., A.H.B.) using the Parkinson’s Disease–Cognitive Rating Scale (PD-CRS), a validated and recommended Level-I tool for cognitive assessment in PD^56^. All study participants were receiving L-dopa and/or dopaminergic agonists (DA). Current medications and dosages were used to calculate the L-dopa daily dose, DA-equivalent L-dopa daily dose, and total L-dopa equivalent daily dose (LEDD)^57^. Participants were required to have received stable dopaminergic treatment for at least 12 weeks and to show a stable clinical response. Motor status and disease stage were assessed using the MDS-UPDRS and the Hoehn and Yahr scale^58^. Exclusion criteria included the presence of motor fluctuations in response to L-dopa, moderate-to-advanced PD (Hoehn and Yahr stage > 2), or cognitive performance in the dementia range (PD-CRS total score < 64)^56,59^. Patients with abnormal blood tests, uncompensated systemic disease, a history of traumatic brain injury, or current use of antipsychotic medication were also excluded.
The presence and type of minor hallucinations were assessed using a two-step procedure. First, item 1.2 (“Hallucinations and Psychosis”) from Part I of the MDS-UPDRS was administered as a standardized and widely accepted screening tool for detecting hallucinatory phenomena in PD^58^. Second, all participants underwent a semi-structured clinical interview developed and routinely employed in our Movement Disorders Unit, designed to explore in detail the phenomenology and frequency of hallucinatory experiences. Only phenomena reported as occurring at least once per week during the past month were considered for classification^12^.
As there is currently no gold standard for the assessment of visual hallucinations in PD, and most existing questionnaires^60,61^ rely primarily on informant information and do not capture phenomenological aspects in sufficient depth^11^, ad hoc semi-structured interviews are commonly used in both clinical and research settings to ensure a more comprehensive evaluation. Accordingly, our interview was informed by established clinical frameworks and prior phenomenological descriptions of hallucinations in PD^3,12,55^, and it was conducted by movement disorder neurologists (J.P., H.B.K., and J.M.L.) with extensive experience in the clinical assessment of hallucinations and related neuropsychiatric symptoms in PD.
The interview systematically addressed a range of perceptual experiences -including visual, auditory, olfactory, and tactile phenomena- to ensure that complex or psychotic-level hallucinations were properly excluded. The occurrence of non-visual experiences in our sample was anecdotal and, in all cases, accompanied by typical minor visual hallucinations such as presence, passage, or pareidolic experiences. These mild cross-modal phenomena have been previously reported in the literature^11,62^ and are considered part of the same continuum of perceptual alterations defining minor hallucinations in PD^4^. The interview also included targeted questions to assess the presence of delusional ideation of any type, including paranoid thoughts, misidentification syndromes, and reduplicative paramnesia, in order to exclude any participant showing evidence of psychotic-level symptoms. These aspects were evaluated using the same semi-structured clinical interview framework and conducted by the movement disorder neurologists responsible for the patients’ clinical assessment.
Based on this combined assessment, participants were categorized into two primary those reporting minor hallucinations (PD-MH) and those without hallucinations (PD-NH).
A second level of classification was applied to capture the potential interaction between hallucinations and cognitive status. Participants were categorized into four clinical subgroups based on the presence or absence of minor hallucinations and of PD-MCI, as defined by performance on the PD-CRS, using the validated cutoff score for PD-MCI (total score < 82)^63^. Accordingly, participants were classified PD-NH/MCI- (no minor hallucinations, no PD-MCI); PD-NH/MCI+ (no hallucinations, with PD-MCI); PD-MH/MCI– (minor hallucinations, no PD-MCI); PD-MH/MCI+ (minor hallucinations, with PD-MCI). This classification allowed us to explore how these two clinical dimensions may independently or jointly shape neurophysiological and behavioral profiles.
All procedures were conducted in accordance with the Declaration of Helsinki and were approved by the Clinical Research Ethics Committee of Sant Pau Hospital. All participants provided written informed consent prior to their inclusion in the study, after receiving a detailed explanation of the study’s aims, procedures, and potential risks.
The visual categorization task was adapted from previous literature^47^. Supplementary Fig. S1 shows a diagram of the procedure. The task comprised a total of 105 distinct visual stimuli, including neutral faces (n = 35), face-like objects (n = 35), and inanimate objects (n = 35). Neutral face images were obtained from the NimStim Face Stimulus Set^64^, while face-like object images were taken from the François & Jean Robert FACES book, as used in prior studies^47,51^. Object stimuli were selected to closely resemble the face-like object images in size, shape, and contour distribution, ensuring that all categories were comparable in overall perceptual structure.
After image editing, all stimuli were converted to black and white and standardized for luminance, contrast, and size (7 × 9 cm) to minimize low-level visual differences. In addition, object images were matched to the face-like object stimuli for contour density, spatial frequency, and internal structural detail, yielding broadly equivalent levels of visual complexity across categories. This matching procedure followed the same validation criteria reported in previous studies, where these stimuli elicited comparable N170 amplitudes in healthy controls, supporting their similarity in perceptual complexity^47^.
Importantly, the paradigm was designed to encompass progressively increasing perceptual and semantic complexity across stimulus, in order to engage successive stages of visual and conceptual processing reflected by distinct ERP components (P100, N170, N300, P600). Although it is inherently difficult to control all aspects of visual complexity all participants were exposed to the same set of stimuli and randomization order. Therefore, any residual variability in perceived complexity was uniformly distributed and is unlikely to have influenced group-level differences.
Participants were instructed to categorize each stimulus as either “looks like a face” or “does not look like a face” as quickly and accurately as possible, responding via button press. To control for lateralization effects, the button assignment for each response option was counterbalanced across participants. Prior to the experimental session, participants received a detailed explanation of the task procedure, including examples of each stimulus category. It was explicitly clarified that only real human faces should be categorized as “looks like a face,” whereas objects and face-like objects should be categorized as “does not look like a face.” To ensure full comprehension, a short training block with feedback was administered before the main task, allowing participants to familiarize themselves with the response mapping and to receive corrective feedback when necessary.
Each trial began with a fixation cross presented at the center of the screen for 600, 700, or 800 ms (randomly determined on each trial, except for the first trial, which was fixed at 700 ms). The stimulus was then displayed centrally on a white background for 700 ms, followed immediately by the next trial. The three stimulus types were presented in randomized order, with each image repeated four times, resulting in a total of 420 trials. The task was divided into blocks of 10 stimuli, with short breaks between blocks to allow participants to rest and blink, thereby minimizing ocular artifacts. Task presentation, timing, and synchronization were implemented using Presentation® software (Version 0.70, www.neurobs.com).
The electroencephalogram (EEG) was recorded using BrainVision Recorder v.1.22 (Brain Products GmbH) with 19 scalp electrodes arranged according to the international 10-20 system, at a sampling frequency of 250 Hz. Electrode derivations included Fp1/2, F3/4, F7/8, C3/4, P3/4, T3/4, T5/6, O1/2, Fz, Cz, and Pz, referenced to the tip of the nose. Two pairs of electrooculogram (EOG) electrodes were used to detect ocular horizontal electrodes (HEOG) were placed at the outer canthi to detect horizontal eye movements, while vertical electrodes (VEOG) were placed beneath the eyes in a bipolar configuration referenced to a central electrode, monitoring vertical eye movements. Electrode impedances were maintained below 5 kΩ to ensure optimal signal quality.
Signal processing involved applying a zero-phase shift 2nd-order Butterworth IIR filter with cutoff frequencies of 0.1 and 40 Hz, along with a 50 Hz notch filter. Independent Component Analysis (ICA) was applied to remove components linked to ocular artifacts, including blinks and lateral eye movements. Component selection was guided by topographical patterns, evaluation of the effect of removing each component on the signal, and identification of the most affected channels^65^.
EEG signals were subsequently segmented into 1000 ms epochs spanning 100 ms before and 900 ms after stimulus onset. Ocular correction was further refined using the Gratton & Coles algorithm to minimize residual eye movement effects^66^, and baseline correction was applied by subtracting the mean activity during the 100 ms prestimulus period from each epoch.
Artifact rejection was based on automatic detection followed by visual verification. Epochs showing voltage steps >20 µV/ms or amplitudes exceeding ±80 µV in any of the recorded channels were automatically flagged and subsequently inspected. Segments confirmed as artifactual were removed, and channels showing artifacts in more than 20% of epochs were considered bad and excluded from further analysis. Participants with fewer than 60% valid trials in any condition were excluded from further analyses.
ERPs were computed by averaging artifact-free epochs for each participant and condition at predefined electrode sites known to capture the maximal amplitude of each O1/O2 for the P100, T5/T6 for the N170, Fz for the N300, and Pz for the P600^27,30–32,39^. Peak-to-peak amplitude was used for the P100 and N170 to reduce baseline variability and better capture early perceptual responses, whereas mean amplitude was used for the N300 and P600, which exhibit broader and slower deflections^67,68^. This approach balances methodological rigor with component-specific sensitivity and has been widely employed in ERP studies of visual categorization and clinical populations. For the P100, the peak-to-peak amplitude was defined as the difference between the positive peak occurring within the first 150 ms after stimulus onset and the mean amplitude of the 100 ms prestimulus baseline at occipital sites. For the N170, peak-to-peak value was calculated as the difference between the negative peak occurring between 150 and 250 ms post-stimulus and the preceding P100 peak, measured at temporo-occipital sites. For the N300, the mean amplitude was calculated within the 250–350 ms time window at fronto-central sites. For the P600, the mean amplitude was calculated within the 450–750 ms time window at centro-parietal sites.
Standardized low-resolution brain electromagnetic tomography (sLORETA) was used to estimate the intracortical generators of key ERP components identified along the time course of stimulus processing. Specifically, we focused on the N170, N300, and P600 components, which are thought to reflect successive stages of perceptual and cognitive evaluation^69–71^. Given the 19-channel EEG setup, source reconstruction was conducted for descriptive purposes only, to visualize the approximate cortical distribution of activity within each component’s latency window. No inferential statistical contrasts were performed. The cortex was divided into 6239 isotropic voxels (5 mm³), and source current density (CSD) was computed for each voxel using scalp-recorded ERP data and a realistic head model from the Montreal Neurological Institute (MNI), with the solution space constrained to cortical grey matter^72^. Source localization was performed for each component within its respective latency window. Given the primarily descriptive aim of this approach, no statistical contrasts were conducted. Instead, source reconstruction was employed qualitatively to visually approximate the cortical distribution of neural generators associated with each processing stage, and to explore the temporal patterns of co-activation across distinct cortical regions.
Clinical and demographic data were analyzed using independent two-tailed t-tests and ANOVA with Tukey’s HSD post-hoc test for continuous variables with normal distribution. For non-normally distributed variables, the Kruskal–Wallis rank-sum test with Conover–Iman post-hoc comparisons was applied. Categorical variables were analyzed using the χ² test. Data are expressed as mean (standard deviation, SD) or median (interquartile range, IQR). Statistical results are reported as t-values for Student’s t-tests, F-values for ANOVA, and H-values for Kruskal–Wallis tests.
To examine behavioral performance, separate repeated-measures ANOVA models were conducted for accuracy and reaction time (RT). In these analyses, stimulus type (faces, face-like objects, objects) was treated as a within-subject factor, and group (PD-MH vs. PD-NH) or clinical subgroup (PD-NH/MCI–, PD-NH/MCI+, PD-MH/MCI–, PD-MH/MCI+) as a between-subject factor, equivalent to a 2 × 3 factorial design. Main effects and interactions (stimulus type × group) were tested, followed by Bonferroni-corrected post hoc pairwise comparisons when appropriate.
Before ERP analyses, the number of artifact-free EEG epochs retained after preprocessing was computed for each condition (faces, face-like objects, objects) to verify data quality and balance across groups. Mean (±SD) trial counts were calculated for each condition and group, and one-way ANOVAs were performed to assess potential differences in the number of valid trials prior to ERP averaging.
For electrophysiological data, repeated-measures ANOVAs were performed separately for each ERP component (P100, N170, N300, and P600). Each model included stimulus type as a within-subject factor and group as a between-subject factor. Analyses were restricted to predefined electrode sites based on prior literature regarding component P100 at occipital electrodes, N170 at T5/T6, N300 at Fz, and P600 at Pz to ensure consistency with previous ERP studies on visual categorization^24,31,38^.
Given the number of statistical comparisons across ERP components, stimulus conditions, and group contrasts, as well as multiple behavioral outcome measures, all p values reported for repeated-measures ANOVAs and post hoc comparisons were Bonferroni-corrected for multiple comparisons. Only p values from single descriptive contrasts (e.g., demographic or clinical variables) are reported uncorrected. All statistical analyses were performed using SPSS software, version 29.0 (IBM Corp., Armonk, NY, USA).
Supplementary Information