Chronic nonspecific low back pain (cNLBP) is a prevalent global health concern. Radiomics enables the extraction of high-dimensional quantitative features from medical images and has shown promise in disease diagnosis, prognostic assessment, and therapeutic response evaluation. To construct and validate an artificial intelligence (AI)-based evaluation model for clinical symptoms in cNLBP patients, leveraging both clinical and radiomics features. The clinical utility of this approach was evaluated in identifying patients at high risk for severe pain. A total of 148 patients with cNLBP were enrolled and stratified by VAS into mild and severe pain groups. Radiomics features from the paraspinal muscles were extracted from lumbar MRI scans. Multiple AI algorithms were applied to construct evaluation models. Logistic regression was used to construct clinical models, radiomics models, and combined clinical - radiomics models, respectively, to compare the predictive power of different models. Model performance was evaluated by multiple methods. Fat infiltration rate of multifidus muscles as significant predictors of pain intensity. The Bagging decision tree model and random forest model achieved higher area under the ROC curve (AUC) values and F1 scores, respectively, in radiomics models. The combined models integrating radiomics and clinical features further increased AUCs. AI algorithms have a significant advantage over traditional algorithms in improving the performance of predictive models. Integrating radiomics features with clinical variables significantly enhances the predictive performance for pain intensity in cNLBP. Multimodal data integration compensates for the limitations of single-modality models, improving both accuracy and stability. This study facilitates early risk stratification of cNLBP patients in clinical practice, enabling clinicians to prioritize intervention for high-risk individuals and optimize the allocation of medical resources. Moreover, the validated high-performance AI models and the multimodal integration strategy lay a foundation for the development of clinical auxiliary tools. Such tools can be integrated into existing clinical workflows to assist clinicians in accurately identifying patients with severe pain at high risk, thereby supporting early intervention and personalized treatment decision-making.
Suo et al. (Thu,) studied this question.