A systematic search was conducted in seven electronic databases from the inception of the databases to December 22, 2024, including PubMed, EMBASE, Cochrane Central Register of Controlled Trials (CENTRAL), Web of Science, China National Knowledge Infrastructure (CNKI), Chongqing VIP (VIP), and Wanfang Database (WF). All the publications within this date were searched without any restrictions of country or article type. Briefly, the general search terms included the following words: (“low back pain” OR “dorsalgia” OR “back disorder*” OR “coccydynia” OR “sciatica” OR “back muscles”) AND (“magnetic resonance imaging” OR “MRI” OR “functional MRI” OR “fMRI” OR “voxel wise” OR “voxel” OR “regional homogeneity” OR “ReHo” OR “amplitude of low frequency fluctuations” OR “ALFF” OR “fractional amplitude of low frequency fluctuation” OR “fALFF” or “mean normalized low-frequency amplitude” OR “mALFF”) AND (“resting state”). The search strategy was transformed to adapt to Chinese electronic databases. Additionally, to identify any potential studies that were not found in our searches, the reference lists of relevant articles and reviews were consulted as additional searches.

This meta-analysis adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines and checklist (Page et al., 2021). The inclusion criteria were as follows: (1) original research published in English or Chinese journals and peer-reviewed; (2) the differences of brain functional activity were compared between adult patients with CLBP and HCs; (3) the analysis methods used in researches was ReHo or ALFF, fALFF, mALFF or included all of them; (4) the standardized three-dimensional coordinates (x, y, z) of results were reported in Talairach space or Montreal Neurological Institute (MNI), or reported null findings; (5) magnetic strength of the MRI scanner was at least 1.5 Tesla.

The exclusion criteria were as follows: (1) researches only reported region of interest (ROI) results; (2) article types were animal studies, case reports or conference proceedings; (3) the number of participants was less than 10 in either the patient group or control group. The detailed screening process is presented in Figure 1. Article screening was performed independently by two authors (XC, PL) and supervised by another reviewer (SC).

If two or more studies used the same data source, only the articles with the largest sample size and more comprehensive information would be included. Besides, for longitudinal or intervention studies, we only included the baseline comparison between patients and HCs.

The information of included articles was independently extracted by two authors (XC and NC). Information of each included study was extracted and presented in Table 1, including: first author, year of publication, demographics (sample size and mean age), illness duration, visual analog scale (VAS) scores, and MRI technical details (MRI scanner, FWHM, analytic methods, correction, threshold, etc.). Statistically significant coordinates in each study were independently extracted according to the requirements of AES-SDM. If coordinates were missing or statistical values were not reported, the corresponding authors of the studies would be contacted by email. If peaks or statistics were not provided from studies reporting null findings, we recorded as none. Any disagreement was discussed by the authors’ group and the result was agreed upon.

The quality of the included studies was assessed based on a 10-point checklist which was used in many previous neuroimaging meta-analyses (Supplementary Tables S1A, S1B) (Guo et al., 2024; Yan et al., 2023). The checklist focused on sample size, demographic and clinical characteristics of the study populations, imaging specific methodology and the quality of the reported results. According to the completeness of published studies, this checklist was scored as full, partial or unmet, with each item receiving a score of 1, 0.5 or 0, respectively. The total score was 10 points. The assessment of each study was carried out independently by two researchers (XC and NC) in a standardized manner. For any differences, XC and NC gave their input and decided with consensus from the third researcher (ZL).

AES-SDM software¹ was used to analyze regional changes in spontaneous brain activity between patients with CLBP and HCs (Radua et al., 2012). Z-values appeared in studies were converted into T-statistics using AES-SDM official online converter.² Above all, the effect-size and variance maps were recreated by combining the extracted peak coordinate information with a Gaussian sliding kernel (full-width at half-maximum = 20 mm). Next, the mean random effects were generated by calculating the study maps voxel-wise based on intra-study variability, sample size, and inter-study heterogeneity (Cheng et al., 2020). The statistical significance and default threshold parameter were set as follows: voxel threshold: p ≤ 0.005, peak height threshold: peak Z ≥ 1.000, cluster extent threshold ≥10 voxels (Radua et al., 2012). Finally, the meta-analysis effect-size map was analyzed by comparing it with a null distribution created by a permutation algorithm.

We initially conducted two separate meta-analyses for the ALFF and ReHo studies using the AES-SDM software. Subsequently, we integrated the functional group data (ALFF + ReHo) to perform another meta-analysis. For all three meta-analyses, we employed a meta-regression analysis weighted by sample size and within-study and between-study variances to assess the correlation between brain changes and clinical variables (age, disease duration, VAS scores). A leave-one-out jackknife sensitivity analysis was used to test the repeatability and robustness of these three results. The approach involves conducting multiple iterations of the same meta-analysis, each time excluding one different dataset. Ultimately, we examine the frequency with which a particular result is consistently retained across these repeated iterations. If a brain region remains significant in all or most of (≥ 50%) the combinations of studies, it can be regarded as highly replicable (Radua et al., 2012; Long et al., 2022). The same statistical significance and default threshold parameter was set as in the main analysis. Egger tests were used to detect the asymmetry of funnel plots for publication bias, where p < 0.05 suggested significant publication bias. Concurrently, akin to the simple overlay of meta-analytical maps in individual meta-analyses, we utilized the multimodal meta-analysis overlap feature within the software to examine the co-localization of functional abnormalities from the first two meta-analyses, taking into account the error in p-values, in order to assess the convergence of results across different modalities (Radua et al., 2012, 2013). This multimodal meta-analysis provides a more statistically robust overlap.

As shown in Figure 1, a total of 1,713 articles were searched. Twenty-three potential eligible articles were screened in this study (Zhang et al., 2017, 2019; Zhou et al., 2018, 2019, 2024a,b; Sun, 2018; Chen et al., 2011; Zou et al., 2019; Hu, 2020; Zhu et al., 2020; Zhang, 2021; Gao et al., 2020; Zhao, 2021; Li et al., 2019; Zhao, 2016; Tan et al., 2019; Wen et al., 2022; Xu et al., 2023; Mei et al., 2024; Zuo, 2024; Chen, 2024; Tang et al., 2023). Due to only one literature on fALFF and two literatures on mALFF being retrieved, we did not conduct separate fALFF/ mALFF meta-analysis of CLBP studies. In addition, 5 articles conducted both ReHo and ALFF analysis (Zhang, 2021; Li et al., 2019; Xu et al., 2023; Mei et al., 2024; Zuo, 2024), but in one of the studies, multi band was extracted for the ALFF analysis (Zhang, 2021). So, we considered these five articles as 9 separate studies for the meta-analysis. At last, 20 articles were included in this meta-analysis (24 studies, 12 ReHo studies, 12 ALFF studies), including 653 patients with CLBP and 653 HCs. Only one study reported no significant between-group differences in brain regions (Li et al., 2019). There was no significant statistical difference in age or gender ratio between patients and HCs. The demographic characteristics, clinical variables, and technical details of these studies were shown in Table 1.

As illustrated in Table 2 and Figure 2, six peak foci were revealed according to the ReHo meta-analysis. Patients with CLBP had increased functional activity in the left inferior temporal gyrus (ITG, z = 2.261, p < 0.0001), left superior frontal gyrus (SFG), medial orbital (z = 1.962, p = 0.0001), right middle frontal gyrus (MFG, z = 1.603, p = 0.0014), right precuneus (z = 1.557, p = 0.0019) and decreased functional activity in the right fusiform gyrus (z = −2.162, p = 0.0001) and bilateral postcentral gyrus (PoCG, z = −1.991, p = 0.0005) compared with HCs.

Patients with CLBP had increased functional activity in the right rolandic operculum (extending to the right insula and right IFG) (z = 3.135, p < 0.0001), left ITG (z = 2.138, p = 0.0008), left middle occipital gyrus (MOG, z = 1.767, p = 0.0033), as well as decreased functional activity in the left paracentral lobule (z = −1.880, p = 0.0005), left postcentral gyrus (PoCG, z = −1.790, p = 0.0008) and bilateral cuneus cortex (z = −1.890, p = 0.0004) compared with HCs (Table 3; Figure 2).

On the basis of the above meta-analyses, we utilized the multimodal functionality of the ASE-SDM software to overlay regions with altered ReHo and ALFF, and discovered increased functional activity in the left ITG of patients with CLBP.

The results of the functional group meta-analysis (ReHo + ALFF) (Table 4; Figure 2) revealed that compared with HCs, patients with CLBP displayed new decreased functional activity in the left inferior parietal (excluding supramarginal and angular) gyri (z = −2.250, p < 0.0001), left median cingulate / paracingulate gyri (z = −1.936, p = 0.0007), right thalamus (z = −1.833, p = 0.0012), right precentral gyrus (PreCG, z = −1.876, p = 0.0010) and right lingual gyrus (z = −1.664, p = 0.0031).

All meta-analyses revealed significant heterogeneity in the right ITG. Egger tests demonstrated that there was no publication bias (p > 0.05) in the meta-analysis of functional data (Supplementary Figures S1A, S1B).

After reading the SDM software instructions and reviewing some meta-analyses related to neuroimaging, it became evident that a consensus regarding the threshold for sensitivity analysis to assess the robustness of the primary outcome is lacking (Long et al., 2022; Wang et al., 2020). Finally, we opted to employ a threshold of 50% in our meta-analysis (Radua et al., 2012; Zhou et al., 2017). The jackknife analysis revealed that all of the aforementioned regions exhibited stability, being reliably detected across the three meta-analyses (Supplementary Tables S2–S4).

As depicted in Tables 5–7, in meta-regression analyses, we examined the potential correlations between clinical variables (age, disease duration, and VAS scores) and these abnormal brain regions. The regression analyses based on ReHo and ALFF revealed correlations between the age of CLBP patients and specific brain regions such as the right precuneus, left PoCG, and left MOG. Consistently across all three regression analyses, a correlation was observed between the ITG and patients’ VAS scores. Furthermore, in both the standalone ALFF regression analysis and the analysis of the integrated functional dataset, correlations were identified between the disease duration of patients and several brain regions, including the PoCG, MOG, ITG, MFG, and right lingual gyrus. Since the datasets were insufficient, we did not perform meta-regression analyses on other continuous variables, such as gender, BMI, medications, and so on.

This study revealed that patients with CLBP exhibit decreased brain activity in the PoCG and thalamus when compared to HCs. Pain is perceived as an imbalance between the ascending pathway and the descending pathway (De Ridder et al., 2022). The somatosensory cortex (SCC), as expressional central of the lateral “painfulness” pathway, encodes the discriminatory/sensory components of the pain. The PoCG is part of the SCC, which primarily governs somatic perception and serves as the ultimate central hub for all sensory pathways in the parietal lobe. The thalamus is considered a crucial relay station closely associated with pain perception and the descending pain modulation pathways. The thalamus projects to the insula and periaqueductal gray matter (PAG), receiving pain signals and transmitting them to the cerebral cortex, and it releases various neurotransmitters in the descending modulation pathways that affect the transmission of pain signals in the spinal cord. Additionally, regression analysis has found a correlation between the PoCG and the duration of CLBP. When lower back pain becomes chronic, the somatic perception dominated by the SCC also undergoes changes, potentially becoming an adaptive process. A recent literature review conducted for a clinical trial targeting the affective pain circuit found that the thalamus is the preferred target for most chronic pain patients during deep brain stimulation (Keifer et al., 2014; Nüssel et al., 2022). Additionally, a study assessing the feasibility and safety of a novel neurofeedback technique based on electroencephalography (EEG ISF-NF) in retraining activity within the SCC and pregenual anterior cingulate cortex (pgACC), as well as dorsal anterior cingulate cortex (dACC), has been confirmed as a safe and effective method for treating CLBP (Adhia et al., 2023). These findings suggest that patients with CLBP exhibit sensory abnormalities in the “pain” pathways and may imply the formation of an adaptive mechanism.

Abnormalities in the prefrontal cortex (PFC), precuneus and rolandic operculum are other important findings in our results. The SFG and MFG, located in the PFC, belong to the descending pain modulatory system and are also the key node of the central executive network. The PFC not only participates in advanced cognitive and memory functions, but also responsible for emotional modulation and stress perception. The role of the PFC in pain regulation has been proved repeatedly in both clinical and pre-clinical work (Ong et al., 2019; Ma et al., 2023; Zhang et al., 2023b). This system has been clearly established to not only inhibit but also facilitate pain (Wilder-Smith, 2011). In the descending pain modulation pathway, the PFC sends descending fibers to the PAG. The functional activity variations of the PFC directly affect the PAG and play a crucial role in pain modulation. A certain review summarized the voxel-based morphometry results of 15 types of chronic pain and showed that changes in the cerebral cortex, particularly in the PFC, occur in chronic pain conditions (May, 2011). Notably, both repetitive transcranial magnetic stimulation (rTMS) and repetitive transcranial direct current stimulation (rTDS) therapies have identified the dorsolateral PFC as a therapeutic target for pain management in clinical practice (Lefaucheur et al., 2020; Li et al., 2024; Che et al., 2021). Furthermore, substantial evidence highlights the pivotal role of the PFC in the development and perpetuation of CLBP.

Precuneus is a part of both the default mode network (DMN) and a key regions of brain activity during resting states (Buckner and DiNicola, 2019; Zhang et al., 2016). Precuneus is involved in higher cognition. DMN is a collection of brain regions that are implicated in various ‘high-level’ cognitive processes. The chronic pain and suffering from CLBP lead to the formation of adaptations and are regulated through connections between the SCC and the DMN. A substantial body of evidence from previous articles indicates that the condition of CLBP leads to functional abnormalities in multiple brain regions within the DMN, further supporting the involvement of the DMN in the emotional processing and cognitive functions associated with chronic pain. Additionally, some studies have utilized the resting-state functional connectivity of the DMN as a predictive marker for assessing the response of CLBP patients to treatment regimens, thereby establishing quantifiable benchmarks for selecting appropriate therapeutic strategies (Tu et al., 2019). Regression analysis has identified a correlation between the precuneus and age, implying a potential decline in cognitive and memory functions as age advances. Studies have demonstrated that older adults with elevated superiority illusion scores tend to have greater gray matter volume in the right precuneus and enhanced FC compared to younger and middle-aged individuals (Shidei et al., 2025). The precuneus undoubtedly assumes a complex and pivotal role in the context of CLBP.

The operculum, defined as the multimodal cortex adjacent to the insula, has a widespread connectivity. Its complex functions include autonomic, cognitive processing, sensory and motor (Mălîia et al., 2018). These functions all participate in the medial pathway. The medial “suffering” pathway, consisting of the rostral to dorsal anterior cingulate cortex and anterior insular cortex, processes the affective motivational aspect of pain. The central pathway’s ability to handle pain in patients with CLBP changes and this “suffering” sensation is further modulation by the central executive network and salience network. This upward pathway overlaps with those two networks, ultimately leading to emotional disorders such as anxiety and depression. As a crucial hub for the integration of emotion and sensation, the insula participates in the transmission and modulation of nociceptive information through its extensive connections with other brain regions. Within the descending modulatory pathway, the insular cortex is connected to multiple brain areas, including the amygdala, anterior cingulate cortex, and PAG, which are key nodes in the transmission and modulation of pain signals. For instance, the connection between the insular cortex and the PAG is involved in the descending modulation of pain. Additionally, the insular cortex engages in the emotional and cognitive aspects of pain through its connection with the anterior cingulate cortex. Studies applying fMRI on patients in pain have found that the insular cortex is one of the most frequently activated brain regions (Apkarian et al., 2005). In short, the degree of pain and the regulation of negative emotions in patients with CLBP are closely related to functional changes in the insular cortex (Labrakakis, 2023).

Additionally, this meta-analysis also found that patients with CLBP have reduced functional activity in the PreCG, fusiform gyrus, precuneus, and lingual gyrus, while there is increased functional activity in the ITG and MOG. Our multimodal analysis also identified abnormalities in the ITG. The PreCG is more anatomically associated with movement and is part of the primary motor cortex (M1), serving as a higher center for human motor function. Still, some proprioceptive fibers also project to the PreCG which encodes the intensity, location, and duration of pain. A recent update on the guidelines for the therapeutic use of rTMS indicates that the application of high-frequency—rTMS to the M1 on the side of pain in patients has a definite analgesic effect (Lefaucheur et al., 2020; Olechowski et al., 2023). The fusiform gyrus is closely associated with facial, bodily, and high spatial frequency stimuli. The lingual gyrus and MOG are also located in the occipital visual cortex, and together with the ITG, they are responsible for visual memory and visual processing. These regions send sensory information through sensory afferents to pain regulation systems, such as the thalamus and amygdala. There is growing evidence that the visual cortex has been participated in the processing of pain (Bush et al., 2021; Wei et al., 2019). In all the results of the regression analysis, a significant correlation between the ITG and VAS scores was found, indicating that functional changes in the ITG play an important role in the perception of pain intensity in patients with CLBP.

This meta-analysis has certain limitations. Firstly, the sample size of the study is relatively small, which leads to insufficient available data. While Müller et al. (2018) have noted that the number of experiments required for a meta-analysis largely depends on the expected effect size, analyses based on small samples should still be approached with caution. Secondly, many differences existed between imaging processing techniques and statistical thresholds as well as the demographic characteristics of patients, publication bias, etc. Moreover, the AES-SDM analysis is based on aggregated data, specifically the coordinates from published studies, rather than original data. This approach may result in less accurate findings. All of these factors may lead to biased results. For those reason, jackknife sensitivity analysis was used to evaluate the reliability and robustness of the results. Despite conducting heterogeneity analysis and Egger’s test, with the p-value of Egger’s test indicating no significant risk of bias, the funnel plot is not symmetrical. When there are significant differences in methodology, patient population, intervention measures, and other aspects, the funnel plot may become asymmetric, even in the absence of publication bias. Greater caution should be exercised when interpreting the results. Thirdly, due to the limited availability of relevant treatment studies, we did not conduct a meta-analysis between abnormal brain regions and studies related to therapeutic effects. The research advancements in related fields are worth further exploring in the future. Finally, this study focused solely on ReHo, ALFF, fALFF and mALFF. More articles using other data analysis methods are not included. Also, the studies included in this meta-analysis were case–control studies rather than longitudinal studies, and they only provided functional data. Any existing structural differences might potentially result in functional brain differences. Our study cannot entirely elucidate that the symptoms of CLBP might either cause brain changes, or on the contrary start from pre-existing brain spontaneous abnormalities. Further integration and analysis of brain structural and functional data and longitudinal studies can be conducted in the future to get further exploration of neuropathological mechanisms of chronic pain, and clarify the causal relationships between brain function alterations and the occurrence of CLBP.