Chronic low back pain (CLBP) is a leading cause of global disability, affecting 568 million patients and resulting in substantial healthcare costs and functional impairment. While virtual reality (VR) has shown promise as a non-pharmacological intervention, most systems rely on fixed routines lacking responsiveness to individual physiological states. Integrating machine learning with real-time biosignal processing offers a pathway to personalized adaptive therapy, yet it remains underexplored in CLBP rehabilitation. This study aims to enhance VR systems with electroencephalography (EEG) sensors to decode CLBP-specific neural patterns using machine learning and dynamically adjust rehabilitation. This next-generation VR trainer will enable personalized, real-time therapeutic intervention in clinical and home settings. The VR-EEG system acquired multichannel EEG data from patients during standardized movement tasks. Model development used publicly available data sets, including the PhysioPain Dataset for movement-related pain EEG patterns, the OSF Chronic Pain EEG Dataset for generalized feature extraction, and the Laser-Evoked Potentials Dataset for high-resolution pain response characterization. Annotated data sets—including pain scores, motion parameters, and clinical labels—trained models using classical regression or convolutional neural networks. Feature extraction focused on time-frequency components and functional connectivity metrics associated with pain perception. The trained model drove real-time VR adaptation—modifying task difficulty, embedding biofeedback, or altering engagement—based on instantaneous pain decoding. The machine learning model achieved over 76% accuracy in classifying pain intensity and significantly reduced subjective pain scores post-intervention compared to conventional VR. This real-time VR-EEG platform pioneers personalized responsive rehabilitation for CLBP. By closing the loop between neural pain signals and therapeutic VR content, it enhances engagement, improves self-regulation, and leads to more effective pain management. Future research will refine algorithm accuracy and assess long-term benefits.
Yunjing Zhang (Sun,) studied this question.