Analysis of Morphological Characteristics of Modic Changes in the Lumbar Spine Based on MRI Imaging Omics and Their Association with Low Back Pain

Objective: To analyse the MRI imaging characteristics, morphological features, and association with low back pain in different types of Modic changes (MC) of the lumbar spine. Features in different types of Modic changes (MC) in the lumbar spine and their association with low back pain. Methods: A retrospective analysis was conducted on the clinical data of 124 patients who underwent lumbar MRI examinations and were diagnosed with Modic changes between March 2024 and February 2025 at a certain hospital. Prospective collection of imaging and clinical data was conducted on 30 patients with different types of lumbar Modic changes and low back pain scores during the same period. Pyradiomics was used to extract MRI morphological and radiomics features in Modic changes, followed by Kruskal-Wallis test, Dunn’s test, Mann-Whitney U test, correlation analysis, LASSO regression screening, and validation of differential features. A classification model was constructed using the support vector machine (SVM) algorithm, and heatmap analysis was performed to investigate the correlation between MRI morphological and radiomics features and low back pain scores. Results: Among 154 patients without low back pain, 34 were Modic Type I, 62 were Type II, and 58 patients with Modic Type III. A total of 7 morphological features and 19 radiomics features showed significant differences in mean values among the three Modic groups (P < 0.05). A Modic classification model based on the differential features was constructed using SVM, with an accuracy rate of 98%. In the correlation analysis, ODI scores were positively correlated with the long-to-short axis ratio and surface area-to-volume ratio of morphological features, and negatively correlated with sphericity and flatness (P < 0.05). Additionally, it was positively correlated with the radiomics feature FS_lbp_3D_m1_glszm_ZoneEntropy (r = 0.380, p < 0.05) and negatively correlated with T1_lbp_3D_m2_glszm_SmallAreaLowGrayLevelEmphasis (r = -0.423, p < 0.05) and FS_wavelet_LLH_firstorder_90thPercentile (r = -0.376, p < 0.05). Conclusion: Morphological and radiomics features differ among different subtypes of Modic changes (MC). An automatic classification model constructed based on these differential features demonstrates high accuracy, and key features are significantly associated with the low back pain functional disability index.