Chronic Pain Protective Behavior Detection with Deep Learning

In chronic pain rehabilitation, physiotherapists adapt physical activity to patients’ performance based on their expression of protective behavior, gradually exposing them feared but harmless and essential everyday activities. As rehabilitation moves outside the clinic, technology should automatically detect such behavior provide similar support. Previous works have shown feasibility automatic detection (PBD) within a specific activity. this article, we investigate use deep learning for PBD across types, using wearable motion capture surface electromyography data collected from healthy participants people with pain. We approach problem by continuously detecting an rather than estimating its overall presence. The best reaches mean F1 score 0.82 leave-one-subject-out cross validation. When is modeled per type, achieves 0.77 bend-down, 0.81 one-leg-stand, 0.72 sit-to-stand, 0.83 stand-to-sit, 0.67 reach-forward. This excellent level agreement average experts’ rating suggesting potential personalized management at home. analyze various parameters characterizing our understand how results could generalize other datasets different levels ground truth granularity.