Predicting Trunk Muscle Activity in Chronic Low Back Pain: Development of a Supervised Machine Learning Model

Introduction: Recently, machine learning adoption has significantly increased across various applications, including the prediction of diseases based on a person’s clinical profile. Accordingly, this study develops and evaluates a supervised machine learning method to predict trunk muscle activity in people with chronic low back pain. Materials and Methods: This was a secondary data analysis from a subgroup of people with nonspecific chronic low back pain. The correlation between labeled data and the output data of muscle activity level was measured through surface electromyography. The result showed a good correlation, suggesting the potential utility of this approach in distinguishing individuals with low back pain from pain-free controls. Results: To validate the performance of the developed machine learning, the results were compared with the SPSS software, version 17. The model’s predictive performance was further assessed using various evaluation methods, including the area under the receiver operating characteristics curve. The study’s findings indicate that the model achieved area under the curve values ranging from 0.5 to 0.9 across all muscles and different tasks for people with back pain. In contrast, the pain-free group exhibited area under the curve values between 0.4 and 0.8. Conclusion: The supervised machine learning approach using logistic regression may offer clinically meaningful predictions in defining the differences in trunk muscle activity between individuals with non-specific chronic low back pain and pain-free controls. While the obtained results demonstrate promise, further studies need to enhance the model’s performance and achieve a more accurate estimation of muscle activity levels.