Authors: Mehmet Akif Güler, Ertuğrul Demirdel, İlknur Albayrak Gezer, Alaaddin Nayman, Ezgi Akyıldız Tezcan
Categories: Research, Chronic low back pain, Core stabilization, Lumbar multifidus, Muscle morphology, Fat infiltration, Physiotherapy, MRI, Pain, Functional outcomes
Source: BMC Musculoskeletal Disorders
Authors: Mehmet Akif Güler, Ertuğrul Demirdel, İlknur Albayrak Gezer, Alaaddin Nayman, Ezgi Akyıldız Tezcan
Chronic non-specific low back pain (CNSLBP) is often associated with morphological changes in the lumbar multifidus muscle (LMF), such as reduced cross-sectional area (CSA) and increased fat infiltration, compromising spinal stability and function. Core stabilization exercises aim to enhance neuromuscular control by targeting deep trunk muscles. However, few randomized trials have investigated their effects on MRI-based muscle morphology and clinical outcomes. This study aimed to compare the effects of an 8-week core stabilization exercise program versus conventional physiotherapy on LMF morphology, pain intensity, disability, and core stability in individuals with CNSLBP.
In this two-arm, parallel-group randomized controlled trial, 36 individuals with CNSLBP (> 3 months) were randomized to a Core Exercise Group (CEG, n = 18) or a Conventional Physiotherapy Group (CPG, n = 18). The 8-week intervention included a 4-week supervised clinical phase and a 4-week home-based phase. The primary outcome was the CSA of the LMF, as this parameter was used for the sample size calculation. Secondary outcomes included LMF fat infiltration (Goutallier grade), pain intensity (visual analog scale), functional disability (Oswestry Disability Index), and core stability (Sahrmann test). Assessments were conducted at baseline and post-intervention. We analyzed within-group change, between-group differences, and group × time interactions using mixed ANOVA; when assumptions were violated, we used non-parametric alternatives (including ART ANOVA). Effect sizes were reported.
Thirty-one participants completed the trial (CEG, n = 15; CPG, n = 16). Both groups showed significant improvements in clinical and functional outcomes (p < 0.05). The CEG additionally demonstrated greater CSA increases—especially at L3–L5 levels (p < 0.01)—and more notable reductions in fat infiltration (p < 0.05). Activity-related pain decreased by 4.4 points in the CEG versus 1.8 points in the CPG (p < 0.001). Functional disability improved by 31 points in the CEG compared to 10 points in the CPG (p < 0.001). Core stability increased to a median Sahrmann test level of 4 in the CEG versus 3 in the CPG (p < 0.001).
Core stabilization exercises yielded superior improvements in LMF morphology, pain relief, functional recovery, and trunk control compared to conventional physiotherapy in individuals with CNSLBP. These findings support the integration of core-focused rehabilitation strategies into standard clinical practice for CNSLBP.
ClinicalTrials.gov Identifier: NCT05302349, protocol ID: 2022–695, registered on March 21, 2022.
The online version contains supplementary material available at 10.1186/s12891-025-09433-x.
Low back pain (LBP) is one of the most prevalent musculoskeletal conditions globally and continues to pose a significant public health burden [1]. It affects individuals across all age groups, leading to diminished quality of life and physical limitations. These consequences collectively reduce work productivity and impose a substantial socioeconomic burden [2]. Clinically, over 90% of LBP cases are classified as non-specific low back pain (NSLBP), in which no identifiable structural cause is present [3]. Among these, approximately 15–20% progress to chronic non-specific low back pain (CNSLBP) [4–12], a condition characterized by persistent pain and functional disability [6]. Contributing factors such as impaired neuromuscular control, segmental instability, and psychosocial influences play a critical role in symptom chronicity and recurrence [13, 14].
Exercise-based rehabilitation is widely recognized as an effective non-invasive treatment approach for managing CNSLBP [15]. Among the various exercise modalities, core stabilization and conventional physiotherapy programs are the most frequently utilized in clinical practice. Dysfunction of deep core muscles, particularly the transversus abdominis and lumbar multifidus muscle (LMF), has been linked to reduced spinal stability, impaired postural control, and the chronicity of pain symptoms [16]. Although both exercise strategies share the goal of alleviating pain and improving function, the current literature presents inconsistent findings. Some studies report that core stabilization exercises yield significantly greater improvements in pain relief and disability reduction [17–19], whereas others suggest similar therapeutic outcomes when compared to general strengthening protocols [20, 21].
More recently, a systematic review concluded that paraspinal muscle fatty infiltration is generally not reversible through exercise in individuals with chronic LBP, challenging the assumption that rehabilitation alone can restore muscle composition [22]. In parallel, randomized controlled trials directly comparing motor control–based interventions and general strengthening have reported mixed some trials demonstrated segment-specific hypertrophy and clinical improvement [23], while others showed limited or non-significant changes in intramuscular fat and function [24, 25]. These studies, published after the initiation of the present trial, highlight the ongoing debate and underscore the need for further evidence using detailed imaging outcomes.
The LMF plays a pivotal role in the segmental stabilization of the spine, given its anatomical characteristics, including a large cross-sectional area and a low fiber-to-length ratio [26]. Extending from the L1 to S1 vertebral levels, it comprises segmentally arranged fibers that contribute to fine spinal control and intersegmental support [27, 28]. In individuals with CNSLBP, LMF atrophy and fatty infiltration are frequently observed and are considered hallmark pathological features [29, 30]. Despite the clinical relevance of these structural changes, most existing studies have assessed the LMF at a single vertebral level, thus limiting the generalizability of their findings [31–33]. To address this gap, the present study utilized high-resolution magnetic resonance imaging (MRI) to evaluate structural adaptations of the LMF at five lumbar levels (from L1–L2 through L5–S1). This comprehensive approach offers a more detailed understanding of segment-specific muscular responses to different rehabilitation strategies.
This randomized controlled trial was designed to compare the effects of core stabilization (defined operationally as low-load, motor control–oriented exercises emphasizing deep stabilizers such as the multifidus, progressed toward functional tasks) and conventional physiotherapy exercises on the cross-sectional area and fatty infiltration of the LMF in individuals with CNSLBP. Secondary outcomes included assessments of general, resting, activity-related, and nocturnal pain, as well as functional disability and lumbopelvic stability. We hypothesized that participants undergoing core stabilization exercises would exhibit significantly greater improvements in both MRI-based structural parameters and clinical outcomes compared to those receiving conventional physiotherapy. When this trial was conceived and registered in early 2022, the most recent systematic reviews and randomized controlled trials addressing paraspinal fatty infiltration and motor control–based interventions had not yet been published. Nevertheless, these emerging studies underscore the importance of high-resolution MRI evidence, such as that provided by the present trial.
This study was designed as a prospective, two-arm, parallel-group randomized controlled trial conducted at Selçuk University between July 2022 and January 2023. The trial was prospectively registered on ClinicalTrials.gov (Identifier: NCT05302349). Ethical approval was obtained from the Health Sciences Ethics Committee of Ankara Yıldırım Beyazıt University on February 7, 2022 (Decision No: 01). All participants provided written informed consent prior to enrollment. No amendments were made to the study protocol after trial initiation. All procedures were conducted in accordance with the ethical standards outlined in the Declaration of Helsinki. This trial is reported in accordance with the CONSORT guidelines, and a completed CONSORT checklist is provided as an additional file.
Participants aged 18 to 65 years with a clinical diagnosis of CNSLBP were recruited from the Physical Medicine and Rehabilitation Outpatient Clinic at Selçuk University. Inclusion criteria were as age between 18 and 65 years, symptom duration longer than three months, the ability to attend scheduled sessions regularly, and voluntary participation. Exclusion criteria included the presence of radicular symptoms (i.e., leg pain consistent with nerve root involvement), history of spinal surgery, any recent neurological deficits, systemic disease, cardiopulmonary contraindications, red flag symptoms, pregnancy, regular use of analgesic medication, or concurrent involvement in another structured exercise program. Withdrawal criteria were defined as experiencing serious adverse events, failure to adhere to the intervention protocol, voluntary withdrawal, or requiring analgesic medication during the intervention period.
The required sample size was calculated using G*Power software, based on data reported by Kliziene et al. [34], who observed a between-group effect size of Cohen’s d = 1.05 for changes in the CSA of the LMF. A priori power analysis indicated that a minimum of 15 participants per group would be sufficient to detect statistically significant differences (α = 0.05, 1–β = 0.80). To compensate for potential dropouts, the final sample size was increased to 18 participants per group (total N = 36).
Randomization was performed using a web-based sequence generator (https://www.randomizer.org), which allocated 36 integers into two equal the Core Exercise Group (CEG) and the Conventional Physiotherapy Group (CPG). Allocation was concealed based on the order of participant enrollment. Although the physiotherapist responsible for administering the interventions and collecting clinical outcome measures was not blinded to group assignment, all MRI analyses were performed by a senior radiologist with 15 years of experience in musculoskeletal imaging, including paraspinal muscle assessment. The radiologist was blinded to both group allocation and time point (pre- vs. post-intervention) to minimize assessment bias. To evaluate intra-rater reliability, all CSA and fat infiltration measurements were repeated one week later by the same radiologist. CSA reliability was assessed using intraclass correlation coefficients (ICC, two-way mixed model, absolute agreement), while fat infiltration reliability was assessed using Cohen’s Kappa.
The intervention period spanned eight weeks and consisted of two consecutive a 4-week supervised clinical phase followed by a 4-week home-based phase. During the supervised phase, all participants attended 20 physiotherapy sessions (five sessions per week) at the Department of Physiotherapy and Rehabilitation, Selçuk University Faculty of Medicine Hospital. All supervised sessions were delivered by a physiotherapist with 12 years of clinical experience in musculoskeletal rehabilitation, with specific expertise in core stabilization and strengthening programs. Each session began with 20 min of hot pack application, 20 min of transcutaneous electrical nerve stimulation (TENS; 1–150 Hz, 20–30 mA), and 10 min of continuous-mode therapeutic ultrasound (1 MHz, 1.5 W/cm^2^) [35]. This was followed by a 5-min self-paced warm-up walk [36], and 15–20 min of dynamic stretching [37, 38], prior to engaging in group-specific exercise protocols. Each session concluded with a 5-min cool-down walk and static stretching. The home-based phase involved three sessions per week over four weeks, totaling 12 sessions. Prior to initiating home exercises, participants received one-on-one instruction and an illustrated guide to ensure proper technique. Each session included a 15–20-min dynamic warm-up [39], the assigned group-specific exercises, and a 5-min cool-down consisting of static stretching. Adherence to the home program was monitored weekly through phone calls and self-reported exercise logs [40], The group-specific exercise protocols implemented during both phases are described below.
Participants in the CEG followed a stabilization protocol adapted from Noormohammadpour et al. [41], targeting the lumbar multifidus and transversus abdominis through abdominal hollowing techniques [42]. Each 40-min session comprised four exercises performed in 3 sets of 10 repetitions, with 10-s rests between repetitions and 1–2-min rests between sets [43, 44]. Swiss balls (55, 65, or 75 cm) were selected according to participant height [45]. Exercise progression followed a structured, staged model, with gradual increases in difficulty while avoiding movements that provoked pain [46].
Participants in the CPG performed general strengthening exercises for the spinal, abdominal, hip, and lower extremity muscles, without emphasis on specific core muscle activation [20, 47]. The program was progressed according to standardized levels of intensity and complexity [48].
Detailed descriptions and illustrations of the main exercises for both groups, including progression parameters and home-based components, are provided in Supplementary File 1.
The primary outcome measure was the cross-sectional area (CSA) of the lumbar multifidus (LMF), as this parameter was used for the sample size calculation. CSA was assessed bilaterally at each lumbar intervertebral level (L1–S1) on axial T2-weighted MRI acquired with a 3 T scanner (Skyra, Siemens Healthcare, Erlangen, Germany). After identifying the LMF in the sagittal plane, three axial slices were selected per at the superior endplate, disc midpoint, and inferior endplate. The fascial boundaries of the LMF were manually traced using RadiAnt DICOM Viewer software, and the CSA was calculated as the mean of the three measurements. All CSA values were reported in square millimeters (mm^2^) (Fig. 1).Fig. 1Bilateral LMF CSA measurement on axial T2-weighted MRI using manual fascial tracing
Secondary outcomes included fatty infiltration of the LMF, pain intensity, functional disability, and core stability. Fatty infiltration was evaluated bilaterally at the same lumbar levels using the Goutallier classification system (grades 0–4), which quantifies intramuscular fat content based on MRI appearance [49, 50]. Axial slices centered at the disc midpoint were used to maintain consistency across evaluations. Representative images for each Goutallier grade are presented in Fig. 2.Fig. 2Axial MRI slices illustrating Goutallier grades of LMF fatty infiltration. A Grade no fat; (B) Grade minimal fat streaks; (C) Grade fat < 50% of muscle volume; (D) Grade fat ≈ 50%; (E) Grade fat > 50%
All MRI assessments (primary CSA and secondary fat infiltration) were conducted at baseline and post-intervention. To evaluate intra-rater reliability, all measurements were repeated one week later by the same radiologist, who was experienced in radiology and nuclear medicine.
Pain intensity was measured using the Visual Analog Scale (VAS) under four general pain, resting pain, activity-related pain, and nocturnal pain [51]. Functional disability was evaluated using the Oswestry Disability Index (ODI), a validated questionnaire that quantifies limitations in daily activities associated with low back pain [52]. Core stability was assessed using the Sahrmann Core Stability Test, which evaluates trunk control through progressively challenging lower-limb movement tasks [53]. All secondary outcomes were assessed at baseline and post-intervention by a physiotherapist trained in standardized clinical assessment procedures.
Statistical analyses were conducted using IBM SPSS Statistics for Windows, Version 25.0 (IBM Corp., Armonk, NY, USA). Due to the unavailability of the Aligned Rank Transform (ART) ANOVA in SPSS, Python 3.8 was utilized for these specific computations.
Descriptive statistics were calculated for all variables. For continuous variables, the mean, standard deviation (SD), median, interquartile range (Q1–Q3), minimum, and maximum values were reported. Categorical variables were summarized as frequencies and percentages. The normality of continuous variables was assessed using the Shapiro–Wilk test. Between-group comparisons were performed using the Independent Samples t-test for normally distributed variables and the Mann–Whitney U test for non-normally distributed variables. Categorical and ordinal variables were analyzed using Pearson’s chi-square test. To examine the interaction effect between group (CEG vs. CPG) and time (baseline vs. post-intervention), a two-way mixed ANOVA was applied for normally distributed data. Prior to this analysis, Levene’s test was used to assess the homogeneity of variances; in cases of violation, Welch's ANOVA was performed. For repeated measures, Repeated Measures ANOVA was conducted when the assumption of sphericity—tested by Mauchly’s test—was met. If sphericity was violated, the Greenhouse–Geisser correction was applied. For non-normally distributed or ordinal outcome variables, group-by-time interaction effects were analyzed using ART ANOVA, and within-group comparisons were assessed using the Wilcoxon Signed-Rank Test. Effect sizes were calculated for all statistically significant results. Eta squared (η^2^) values were reported for ANOVA models and interpreted as small (0.01–0.06), medium (0.06–0.14), or large (> 0.14). For Wilcoxon tests, the rank-biserial correlation was reported as the effect size and interpreted based on the thresholds proposed by Goos-Sampson (2018): negligible (r < 0.1), small (r ≈ 0.1), medium (r ≈ 0.3), and large (r ≈ 0.5).
In addition, the individual changes from pre- to post-intervention were calculated and presented as mean ± SD. A two-tailed p-value of < 0.05 was considered statistically significant. No interim analyses or stopping rules were applied during the trial.
A total of 74 individuals were assessed for eligibility between July 2022 and January 2023. Of these, 12 were excluded for not meeting the inclusion criteria, and 26 declined to participate. The remaining 36 individuals were enrolled in the trial and randomly assigned to one of two intervention groups using a web-based randomization tool (https://www.randomizer.org). A pre-generated allocation sequence comprising numbers 1 to 36 was used to ensure equal distribution of participants between the CEG and CPG. As participants were enrolled, they were sequentially assigned to the group corresponding to their number in the randomization list, based on the order of recruitment. Following group allocation, five participants were withdrawn from the study due to non-compliance with the intervention protocol—three from the CEG and two from the CPG. Consequently, 31 participants (CEG, n = 15; CPG, n = 16) completed the study and were included in the final analysis. The complete participant flow throughout the study phases is illustrated in Fig. 3.Fig. 3CONSORT flow diagram depicting participant progress through the study phases
No adverse events were reported in either group during the trial. Participants confirmed that they did not receive any concurrent interventions throughout the study period. Adherence to both the supervised sessions and the home exercise program was complete, as all participants who remained in the study fully complied with the intervention protocol, confirmed through exercise logs and weekly phone calls.
There were no statistically significant differences between the CEG and CPG in terms of age, body mass index (BMI), pain duration, sex distribution, educational level, or employment status (p > 0.05 for all comparisons). These results indicate that the groups were comparable at baseline, ensuring the validity of subsequent between-group analyses (Table 1).Table 1Baseline sociodemographic and physical characteristics of participantsCPG (n = 16)CEG (n = 15)Test statistict, Z, χ^2^pAge (years) Mean ± SD35.88 ± 14.5234.47 ± 14.64Z = –0.180.857 M (Q1–Q3)35 (21–49.5)32 (21–50) min–max20–5820–61BMI (kg/m^2^) Mean ± SD26.39 ± 5.8524.78 ± 4.85t = 0.8330.412 M (min–max)26.02 (17.12–37.72)24.38 (17.85–32.04)Pain duration (years) Mean ± SD2.63 ± 1.592.67 ± 1.68t = 0.0710.944 M (min–max)2.5 (0.5–5)2 (0.5–6)Sex, n (%) Female7 (46.67%)8 (53.33%)χ^2^ = 0.2850.594 Male9 (56.25%)7 (43.75%)Education level, n (%) Primary3 (18.75%)2 (13.33%)χ^2^ = 3.1040.376 High School6 (37.50%)9 (60.00%) University6 (37.50%)2 (13.33%) Postgraduate1 (6.25%)2 (13.33%)Employment status, n (%) Yes9 (60.00%)6 (40.00%)χ^2^ = 0.8190.366 No7 (43.75%)9 (56.25%)CPG Conventional Physiotherapy Group, CEG Core Exercise Group, n Number of participants, SD Standard deviation, M Median, Q1–Q3 Interquartile range (1st–3rd quartiles), min–max Minimum and maximum values, t Independent samples t-test statistic, Z Mann–Whitney U test statistic (standardized), χ^2^ Pearson chi-square test statistic, % Percentage
Intra-rater reliability for CSA measurements was excellent (ICC = 0.999, 95% CI = 0.995–1.000, p < 0.001). Reliability for fat infiltration grading was substantial (κ = 0.75, p < 0.001).
The CSA of the LMF was assessed bilaterally at each vertebral level from L1–2 to L5–S1. At baseline, no statistically significant differences were observed between the groups (p > 0.05). Similarly, post-intervention comparisons revealed no significant between-group differences in the CSA of the LMF at any level. However, within-group analyses indicated significant increases in CSA at all levels for both CEG and CPG (p < 0.05). These increases were more pronounced in the CEG, as reflected by higher F-values and larger effect sizes (η^2^). In the CEG, the most substantial increases in CSA were observed at L4–5 left (η^2^ = 0.859), L4–5 right (η^2^ = 0.857), L3–4 right (η^2^ = 0.821), and L3–4 left (η^2^ = 0.820). In contrast, the CPG showed the greatest changes at L5–S1 left (η^2^ = 0.642), L5–S1 right (η^2^ = 0.598), L3–4 right (η^2^ = 0.592), and L4–5 left (η^2^ = 0.549). A statistically significant group × time interaction effect was observed (p < 0.05), indicating that participants in the CEG experienced greater improvements in the CSA of the LMF over time compared to those in the CPG (Fig. 4). The strongest interaction effects were noted at L4–5 left (η^2^ = 0.538) and L4–5 right (η^2^ = 0.495), while the smallest effects were seen at L3–4 left (η^2^ = 0.225) and L2–3 left (η^2^ = 0.253). Detailed results are presented in Table 2.Fig. 4Between-group comparison of LMF CSA changes (Δx) after interventionTable 2Lumbar multifidus CSA (mm^2^) at each vertebral level before and after the interventionCPG (n = 16)CEG (n = 15)Test statisticMean ± SDM (min–max)Mean ± SDM (min–max)T**pL1-2 Right LMF CSAPre343.94 ± 72.13352.5 (195–496)320.67 ± 80.39293 (216–496)0.8490.403Post345.00 ± 71.51352.5 (196–496)330.27 ± 84.25301 (219–518)0.5260.603FRMF = 4.446p = 0.052η2 = 0.229F = 34.960p < 0.001η2 = 0.714FABF = 26.629p < 0.001η2 = 0.479Δx1.06 ± 2.029.60 ± 6.36L1-2 Left LMF CSAPre343.13 ± 71.09356.5 (199–495)319.47 ± 79.04299 (214–472)0.8770.387Post344.32 ± 70.61357 (201–496)327.27 ± 82.78299 (217–495)0.6180.541FRMF = 6.987p = 0.018η2 = 0.318F = 19.644p < 0.001η2 = 0.584FABF = 14.048p < 0.001η2 = 0.326Δx1.19 ± 1.807.80 ± 6.82L2-3 Right LMF CSAPre480.50 ± 95.42493 (255–673)459.20 ± 92.20429 (348–685)0.6310.533Post482.44 ± 94.87496 (257–672)467.07 ± 93.75438 (355–701)0.4530.654FRMF = 13.069p = 0.003η2 = 0.466F = 40.393p < 0.001η2 = 0.743FABF = 20.203p < 0.001η2 = 0.411Δx1.94 ± 2.147.87 ± 3.11L2-3 Left LMF CSAPre479.87 ± 92.36490.5 (263–668)458.00 ± 89.55446 (342–651)0.6690.509Post482.00 ± 92.16493.5 (264–668)463.87 ± 90.75455 (350–666)0.5520.585FRMF = 11.087p = 0.005η2 = 0.425F = 32.597p < 0.001η2 = 0.700FABF = 9.839p < 0.001η2 = 0.253Δx2.13 ± 2.555.87 ± 3.47L3-4 Right LMF CSAPre703.56 ± 141.84723.5 (439–994)665.67 ± 113.44642 (490–906)0.8180.420Post707.31 ± 141.08725.5 (444–1001)675.00 ± 115.18651 (504–924)0.6960.492FRMF = 21.574p < 0.001η2 = 0.592F = 64.112p < 0.001η2 = 0.821FABF = 15.895p < 0.001η2 = 0.354Δx3.75 ± 3.229.33 ± 3.88L3-4 Left LMF CSAPre692.88 ± 119.42725 (442–857)667.13 ± 109.43654 (496–881)0.6240.537Post696.63 ± 119.04724.5 (445–868)674.87 ± 111.35663 (505–895)0.5250.604FRMF = 14.868p = 0.002η2 = 0.498F = 63.773p < 0.001η2 = 0.820FABF = 8.403p = 0.007η2 = 0.225Δx3.75 ± 3.897.73 ± 3.53L4-5 Right LMF CSAPre995.00 ± 155.141029 (668–1353)946.93 ± 156.50912 (724–1327)0.8580.398Post1000.56 ± 156.461028 (672–1367)965.07 ± 161.62946 (741–1355)0.6210.539FRMF = 17.353p < 0.001η2 = 0.536F = 84.237p < 0.001η2 = 0.857FABF = 28.436p < 0.001η2 = 0.495Δx5.56 ± 5.3418.13 ± 5.58L4-5 Left LMF CSAPre988.13 ± 128.971011 (689–1222)943.53 ± 147.02945 (732–1286)0.8990.376Post992.31 ± 129.621018 (691–1234)958.60 ± 151.13924 (738–1310)0.6680.509FRMF = 18.263p < 0.001η2 = 0.549F = 85.595p < 0.001η2 = 0.859FABF = 33.749p < 0.001η2 = 0.538Δx4.19 ± 3.9215.07 ± 5.21L5-S1 Right LMF CSAPre1155.13 ± 162.971153 (745–1438)1126.67 ± 166.311168 (919–1461)0.4810.634Post1161.94 ± 162.471156 (752–1457)1148.27 ± 172.291068 (936–1499)0.2270.822FRMF = 22.347p < 0.001η2 = 0.598F = 52.687p < 0.001η2 = 0.790FABF = 20.820p < 0.001η2 = 0.418Δx6.81 ± 5.7621.60 ± 10.26L5-S1 Left LMF CSAPre1153.25 ± 152.601162 (775–1403)1117.80 ± 154.761069 (938–1455)0.6420.526Post1159.19 ± 152.191167 (781–1420)1135.80 ± 160.211070 (947–1481)0.4170.680FRMF = 26.865p < 0.001η2 = 0.642F = 48.462p < 0.001η2 = 0.776FABF = 19.005p < 0.001η2 = 0.396Δx5.94 ± 4.5818.00 ± 12.28CPG Conventional Physiotherapy Group, CEG Core Exercise Group, Pre Baseline, Post Post-intervention, t Independent samples t-test, p p-value, FRM Within-group repeated measures ANOVA, FAB Two-way mixed ANOVA (group × time interaction), F F-statistic, η^2^ Eta squared (effect size), Δx Within-group change score (mean ± SD), M Median, SD Standard deviation, min–max Range (minimum–maximum), CSA Cross-sectional area, LMF Lumbar multifidus muscle, L1–2 to L5–S1 Intervertebral levels
The fat infiltration of the LMF was evaluated at each vertebral level from L1–2 to L5–S1. At baseline, no statistically significant differences were found between the groups (p > 0.05). Post-intervention comparisons revealed significant reductions in fat infiltration at all levels except for L1–2 right (p = 0.124), L1–2 left (p = 0.204), and L3–4 left (p = 0.077), as illustrated in Fig. 5.Fig. 5Between-group comparison of LMF fat infiltration changes (ΔGoutallier score) after intervention
Within-group analyses demonstrated statistically significant improvements in the CEG at all vertebral levels (p < 0.05). In contrast, CPG exhibited significant changes only at selected levels, with no meaningful reductions at L1–2 right/left, L2–3 right/left, and L3–4 left (p > 0.05). The CEG showed larger reductions in fat infiltration compared to the CPG, as indicated by higher F-values and greater effect sizes (|r|). The most prominent improvements in the CEG were observed at L5–S1 right (|r|= 0.936), L4–5 left (|r|= 0.915), L4–5 right (|r|= 0.909), and L5–S1 left (|r|= 0.867). For the CPG, the largest changes were found at L5–S1 right (|r|= 0.707), L4–5 left (|r|= 0.662), L5–S1 left (|r|= 0.612), and L4–5 right (|r|= 0.559). Group × time interaction effects were statistically significant at all levels except for L1–2 right (p = 0.161), L1–2 left (p = 0.172), L3–4 left (p = 0.298), and L5–S1 left (p = 0.166). The largest interaction effect was observed at L4–5 right (η^2^ = 0.231), followed by moderate effects at L4–5 left (η^2^ = 0.137), L2–3 left (η^2^ = 0.092), and L2–3 right (η^2^ = 0.084). Small effects were found at L1–2 right (η^2^ = 0.030), L1–2 left (η^2^ = 0.029), L3–4 left (η^2^ = 0.018), and L5–S1 left (η^2^ = 0.027). Detailed results are presented in Table 3.Table 3Lumbar multifidus fat infiltration (Goutallier score) at each vertebral level pre- and post-interventionCPG (n = 16)CEG (n = 15)Test statisticMean ± SDM (Q1-Q3)Mean ± SDM (Q1-Q3)Z**pL1-2 Right LMFPre1.31 ± 0.951 (1–1)1.33 ± 0.621 (1–2)−0.6380.537Post1.25 ± 1.001 (1–1)0.73 ± 0.701 (0–1)−1.5370.124WSTW = 0.00ZW = −1.00p = 0.317*|r|= 0.25W* = 0.00ZW = −3.00p = 0.003*|r|= 0.775ARTF* = 2.014p = 0.161η2 = 0.030Δx−0.06 ± 0.16−0.60 ± 0.51L1-2 Left LMFPre1.25 ± 1.001 (1–1)1.27 ± 0.461 (1–2)−0.8160.414Post1.25 ± 1.001 (1–1)0.80 ± 0.701 (0–1)−1.2710.204WSTW = 0.00ZW = 0.00p = 1.000*|r|= 0.000W* = 0.00ZW = −2.65p = 0.008*|r|= 0.683ARTF* = 1.909p = 0.172η2 = 0.029Δx0.00 ± 0.00−0.47 ± 0.52L2-3 Right LMFPre1.63 ± 0.891 (1–2)1.47 ± 0.521 (1–2)−0.8750.157Post1.44 ± 0.631 (1–2)0.47 ± 0.521 (0–1)−3.723 < 0.001WSTW = 0.00ZW = −1.73p = 0.083*|r|= 0.433W* = 0.00ZW = −3.22p < 0.001*|r|= 0.831ARTF* = 5.608p = 0.021η2 = 0.084Δx−0.19 ± 0.31−1.00 ± 0.53L2-3 Left LMFPre1.50 ± 0.821 (1–2)1.53 ± 0.522 (1–2)−0.6790.497Post1.31 ± 0.871 (1–2)0.73 ± 0.601 (0–1)−2.1160.034WSTW = 0.00ZW = −1.73p = 0.083*|r|= 0.433W* = 0.00ZW = −2.97p = 0.003*|r|= 0.743ARTF* = 6.150p = 0.016η2 = 0.092Δx−0.19 ± 0.40−0.80 ± 0.52L3-4 Right LMFPre1.63 ± 0.891 (1–2)1.60 ± 0.741 (1–2)−0.7730.440Post1.38 ± 0.621 (1–2)0.73 ± 0.591 (0–1)−1.9750.048WSTW = 0.00ZW = −2.00p = 0.046*|r|= 0.500W* = 0.00ZW = −3.13p = 0.002*|r|= 0.807ARTF* = 4.379p = 0.041η2 = 0.064Δx−0.25 ± 0.45−0.87 ± 0.71L3-4 Left LMFPre1.75 ± 0.861 (1–2)1.60 ± 0.741(1–2)−0.4560.648Post1.56 ± 0.891 (1–2)1.00 ± 0.661 (1–1)−1.7710.077WSTW = 0.00ZW = −1.73p = 0.083*|r|= 0.433W* = 0.00ZW = −3.00p = 0.003*|r|= 0.775ARTF* = 1.102p = 0.298η2 = 0.018Δx−0.19 ± 0.40−0.60 ± 0.46L4-5 Right LMFPre2.31 ± 0.602 (2–3)2.33 ± 0.822 (2–3)−0.2280.820Post2.00 ± 0.632 (2–2)0.73 ± 0.701 (0–1)−3.893 < 0.001WSTW = 0.00ZW = −2.24p = 0.025*|r|= 0.559W* = 0.00ZW = −3.52p < 0.001*|r|= 0.909ARTF* = 22.336p < 0.001η2 = 0.231Δx−0.31 ± 0.45−1.60 ± 0.55L4-5 Left LMFPre2.31 ± 0.602 (2–3)2.33 ± 0.822 (2–3)−0.2280.820Post1.88 ± 0.622 (1–2)1.00 ± 0.761 (0–2)−2.9810.003WSTW = 0.00ZW = −2.65p = 0.008*|r|= 0.662W* = 0.00ZW = −3.54p < 0.001*|r|= 0.915ARTF* = 10.741p = 0.002η2 = 0.137Δx−0.44 ± 0.50−1.33 ± 0.51L5-S1 Right LMFPre2.69 ± 0.603 (2–3)2.53 ± 0.642 (2–3)−0.7760.438Post2.19 ± 0.552 (2–3)1.33 ± 0.621 (1–2)−3.480 < 0.001WSTW = 0.00ZW = −2.83p = 0.005*|r|= 0.707W* = 0.00ZW = −3.63p < 0.001*|r|= 0.936ARTF* = 5.964p = 0.018η2 = 0.076Δx−0.50 ± 0.52−1.20 ± 0.51L5-S1 Left LMFPre2.69 ± 0.603 (2–3)2.47 ± 0.742 (2–3)−0.8980.369Post2.31 ± 0.602 (2–3)1.60 ± 0.632 (1–2)−2.8120.005WSTW = 0.00ZW = −2.45p = 0.014*|r|= 0.612W* = 0.00ZW = −3.36p < 0.001*|r|= 0.867ARTF* = 1.967p = 0.166η2 = 0.027Δx−0.38 ± 0.50−0.87 ± 0.54CPG Conventional Physiotherapy Group, CEG Core Exercise Group, Pre Baseline, Post Post-intervention, WST Wilcoxon signed-rank test, ART Aligned Rank Transform ANOVA (group × time interaction), F F-statistic, p p-value, W Wilcoxon test statistic, ZW Standard score of W in normal distribution, Z Mann–Whitney U test statistic (standardized), η^2^ Eta squared (effect size), |r| Rank-biserial correlation, Δx Within-group change score (mean ± SD), M Median, Q1–Q3 Interquartile range (1st–3rd quartiles), SD Standard deviation, L1–2 to L5–S1 Intervertebral levels
Pain intensity, measured using the VAS, showed no statistically significant differences between the groups at baseline across all domains (general, resting, activity-related, and nocturnal) (p > 0.05). Post-intervention comparisons revealed that the CEG had significantly lower pain scores across all domains compared to the CPG (p < 0.05). Within-group analyses indicated significant reductions in pain intensity in both groups (p < 0.05), with greater improvements observed in the CEG. The largest effect was observed in general pain (η^2^ = 0.896), and the most prominent group × time interaction was found for activity-related pain (η^2^ = 0.418) (Table 4).Table 4Between- and within-group comparison of pain ıntensity scores (VAS)CPG (n = 16)CEG (n = 15)Test statisticMean ± SDM (min–max)Mean ± SDM (min–max)T**pGeneral PainPre7.49 ± 1.257.35 (5.4–9.2)7.69 ± 0.957.6 (6–9.2)−0.5150.611Post5.25 ± 1.255.15 (2–7.6)4.30 ± 0.844.2 (2.9–6)2.4670.020FRMF = 54.276p < 0.001η2 = 0.783F = 120.487p < 0.001η2 = 0.896FABF = 7.106p = 0.012η2 = 0.197Δx−2.72 ± 0.44−3.39 ± 0.34Resting PainPre5.46 ± 1.515.95 (2.9–8)5.70 ± 1.605.20 (3.6–9.3)−0.4250.674Post3.63 ± 1.013.4 (2.2–5.5)2.28 ± 0.922 (1.2–4.2)3.872 < 0.001FRMF = 42.551p < 0.001η2 = 0.739F = 81.346p < 0.001η2 = 0.853FABF = 11.471p = 0.002η2 = 0.283Δx−1.83 ± 0.47−3.42 ± 0.49Nocturnal PainPre4.77 ± 2.085 (1.8–8.1)4.74 ± 2.284 (1.8–9)0.3700.971Post3.15 ± 1.413 (1.4–7.3)1.82 ± 0.852 (0–4)3.1450.004FRMF = 27.650p < 0.001η2 = 0.648F = 33.193p < 0.001η2 = 0.703FABF = 4.957p = 0.034η2 = 0.146Δx−1.62 ± 0.65−2.92 ± 0.65Activity-Related PainPre7.69 ± 1.507.9 (4.9–9.7)8.07 ± 0.908 (7–9.4)0.8330.412Post5.86 ± 1.616 (3.6–9)3.62 ± 1.433.4 (2.1–7.3)4.075 < 0.001FRMF = 29.089p < 0.001η2 = 0.660F = 91.385p < 0.001η2 = 0.867FABF = 20.853p < 0.001η2 = 0.418Δx−1.83 ± 0.57−4.45 ± 0.45CPG Conventional Physiotherapy Group, CEG Core Exercise Group, Pre Baseline, Post Post-intervention, t Independent samples t-test, p p-value, FRM Within-group repeated measures ANOVA, FAB Two-way mixed ANOVA (group × time interaction), F F-statistic, η^2^ Eta squared (effect size), Δx Within-group change score (mean ± SD), M Median, SD Standard deviation, min–max Range (minimum–maximum), VAS Visual Analog Scale
Functional disability, measured by the ODI, significantly improved in both groups following the intervention (p < 0.05). However, the CEG exhibited a greater reduction (η^2^ = 0.800) compared to the CPG (η^2^ = 0.685). A significant group × time interaction was observed (F = 21.841, p < 0.001, η^2^ = 0.430), indicating more pronounced functional recovery in the CEG (Table 5).Table 5Between- and within-group comparison of functional disability (ODI)CPG (n = 16)CEG (n = 15)Test StatisticMean ± SDM (min–max)Mean ± SDM (min–max)T**pFunctional DisabilityPre44.21 ± 15.3039 (18–72)52.06 ± 13.5754 (36–84)−1.5080.142Post33.92 ± 14.4431.33 (12–68)21.20 ± 9.1520 (6–36)2.9050.007FRMF = 32.687p < 0.001η2 = 0.685F = 56.007p < 0.001η2 = 0.800FABF = 21.841p < 0.001η2 = 0.430Δx−10.29 ± 5.43−30.86 ± 4.37CPG Conventional Physiotherapy Group, CEG Core Exercise Group, Pre Baseline, Post Post-intervention, t Independent samples t-test, p p-value, FRM Within-group repeated measures ANOVA, FAB Two-way mixed ANOVA (group × time interaction), F F-statistic, η^2^ Eta squared (effect size), Δx Within-group change score (mean ± SD), M Median, SD Standard deviation, min–max Range (minimum–maximum), ODI Oswestry Disability Index
Core stability, assessed using the Sahrmann Core Stability Test, significantly improved in both groups (p < 0.05). The CEG showed a larger effect size (|r|= 0.839) than the CPG (|r|= 0.530), suggesting greater enhancement in trunk control. A statistically significant group × time interaction was also detected (F = 12.653, p < 0.001, η^2^ = 0.123), indicating superior gains in core stability among participants in the CEG (Table 6).Table 6Between- and within-group comparison of core stability (Sahrmann Test)CPG (n = 16)CEG (n = 15)Test StatisticMean ± SDM (Q1-Q3)Mean ± SDM (Q1-Q3)Z**pCore StabilityPre2.63 ± 0.813 (2–3)2.33 ± 0.822 (2–3)−0.9970.319Post3.00 ± 0.903 (2–5)4.20 ± 0.684 (4–5)−3.399 < 0.001WSTW = 15.00ZW = 2.12p = 0.034*|r|= 0.530W* = 91.00ZW = 3.25p < 0.001*|r|= 0.839ART**F* = 12.653p < 0.001η2 = 0.123Δx0.37 ± 0.821.87 ± 0.99CPG Conventional Physiotherapy Group, CEG Core Exercise Group, Pre Baseline, Post Post-intervention, WST Wilcoxon signed-rank test, ART Aligned Rank Transform ANOVA (group × time interaction), F F-statistic, p p-value, W Wilcoxon test statistic, ZW Standard score of W in normal distribution, Z Mann–Whitney U test statistic (standardized), η^2^ Eta squared (effect size), |r| Rank-biserial correlation, Δx Within-group change score (mean ± SD), M Median, Q1–Q3 Interquartile range (1st–3rd quartiles), SD Standard deviation
This study demonstrated that core stabilization exercises yielded superior improvements in LMF morphology, specifically in CSA and fat infiltration, compared to conventional physiotherapy in individuals with CNSLBP. These findings support the hypothesis that targeting deep stabilizing muscles through segmental control-based training leads to structural adaptations in spinal musculature. Notably, the most pronounced increases in the CSA of the LMF were observed at L4–5 and L3–4 levels in the CEG group, suggesting selective hypertrophy associated with targeted activation of deep muscles. The significant reduction in fat infiltration, especially at the L4–5 and L5–S1 levels, indicates not only improved muscle composition but also enhanced neuromuscular efficiency in stabilizing the lumbar spine. These changes are consistent with previous studies reporting the responsiveness of multifidus to stabilization training [31, 32], yet our findings offer a broader segmental analysis across five vertebral levels using 3 T MRI—adding depth and precision to the current literature. However, the interpretation of fatty infiltration changes requires caution. A recent systematic review concluded that paraspinal fatty infiltration is generally not reversible through exercise in individuals with LBP [22]. Against this background, our findings of short-term reductions in LMF fat content may appear unusually strong. Yet, randomized controlled trials have begun to show that exercise interventions can elicit measurable morphological changes. Fortin et al. [23], demonstrated improvements in paraspinal muscle morphology and function after combined motor control and extensor strengthening training, while Rosenstein et al. [24] reported significant short-term reductions in fatty infiltration alongside CSA gains after only 8 weeks of exercise. Moreover, Rosenstein et al. [25] highlighted in a protocol study that aquatic stabilization programs may further enhance spinal and gluteal muscle adaptations. Taken together, these studies indicate that although long-term reversibility of fatty infiltration remains debated, targeted and segment-specific training protocols can produce short-term morphological benefits in selected spinal segments. Notably, all participants in this trial presented with Goutallier grades ≤ 3 at baseline. Therefore, while short-term reductions in fatty infiltration were observed, extrapolation to patients with grade 4 fatty infiltration should be made with caution, as previous studies suggest advanced degeneration may limit the potential for reversal [22].
The reduction in fat content is particularly important as fat infiltration of the LMF is a known correlate of chronicity and functional impairment in LBP [33]. While some prior studies reported site-specific changes (e.g., L5-S1 only), our results reveal bilateral improvements at both L4–5 and L5–S1 suggesting the effectiveness of our 8-week CEG protocol in reversing pathological degeneration. Moreover, using high-resolution MRI provided objective insights into muscle quality, supporting the clinical relevance of LMF morphology as a therapeutic target in CNSLBP.
Regarding pain outcomes, both groups exhibited significant reductions across all pain domains; however, the CEG group experienced greater improvements, particularly in activity-related and general pain, with large effect sizes. These results imply that dynamic stabilization contributes more effectively to controlling nociceptive input during functional activities. The moderate but consistent reduction in nocturnal pain in the CEG group may reflect improvements in passive postural control during rest. The literature supports these meta-analyses and RCTs have reported that stabilization exercises outperform conventional regimens in reducing pain intensity [19, 54], although few studies have investigated pain in activity-specific contexts. Our study addressed this gap by evaluating pain during movement, rest, and nighttime, increasing the clinical applicability of the results.
Functional disability, as assessed by the Oswestry Disability Index, improved significantly in both groups, yet the CEG group again demonstrated superior recovery. An approximately 59% reduction in ODI scores among CEG participants highlights the functional relevance of deep muscle recruitment. This aligns with França et al. [55], and Javadian et al. [56], who emphasized the role of core musculature endurance and neuromuscular coordination in enabling pain-free movement. The conventional group also showed benefit, likely due to generalized strengthening and circulation effects, but lacked the postural specificity necessary for deeper recovery.
Lumbopelvic core stability, assessed through the Sahrmann test, also improved in both groups, with larger gains in the CEG group. This reinforces the notion that activation of the lumbar multifidus provides segmental spinal control. Improved proprioceptive awareness and intra-abdominal pressure regulation likely contributed to these results. The literature offers mixed findings on core stability, with some trials reporting no significant group differences [57], possibly due to heterogeneous populations or shorter intervention durations. Our study included only chronic cases and implemented a consistent protocol, reducing potential confounders.
The findings of this study must be interpreted in light of its limitations. First, the 8-week follow-up period may not capture long-term effects of core stabilization, and the study included only pre- and post-intervention measurements without long-term follow-up, which limits conclusions regarding the persistence of observed effects. Second, the study was conducted in a single center with a homogenous population, limiting generalizability. Third, while the intervention was standardized, individualization based on specific motor control deficits was not performed, which may have influenced responsiveness in some participants. Additionally, no adverse events were reported, suggesting the safety of the protocol; however, larger multicenter trials are needed to confirm both safety and efficacy across diverse settings. Moreover, fatty infiltration was assessed from a single axial slice at each level and graded using the qualitative Goutallier system. This approach may not fully capture the heterogeneous distribution of intramuscular fat along the spine and is less precise than quantitative imaging methods, which should be considered a limitation. Finally, although the transversus abdominis is also a key stabilizer, its quantitative MRI assessment is technically challenging and less standardized; therefore, its exclusion should be considered a limitation. Furthermore, since no participants presented with Goutallier grade 4 fatty infiltration, the generalizability of our findings to individuals with severe fatty degeneration is limited.
Despite these limitations, the study provides robust evidence supporting the short-term efficacy of core stabilization exercises in CNSLBP. By integrating clinical, functional, and imaging-based outcomes, it offers a comprehensive evaluation of therapeutic effectiveness. Importantly, the use of objective imaging outcomes such as MRI-based CSA and fat infiltration places this study at the intersection of clinical and structural rehabilitation research. These results not only support the inclusion of core stabilization protocols in rehabilitation but also suggest that targeting LMF structure may yield measurable functional benefits. Future research should explore long-term adaptations and extend findings to broader populations.
This randomized controlled trial demonstrated that core stabilization exercises are more effective than conventional physiotherapy in improving lumbar multifidus morphology, reducing pain intensity, enhancing lumbopelvic stability, and alleviating functional disability in individuals with chronic non-specific low back pain. These findings underscore the clinical relevance of incorporating core-focused rehabilitation into standard treatment protocols. These results are most applicable to individuals with mild-to-moderate intramuscular fat infiltration (Goutallier ≤ 3); in patients with advanced degeneration (grade 4), morphological reversibility may be less likely. Further research is warranted to evaluate the long-term effects and generalizability of these interventions.
Supplementary Material 1.