Spike train analysis in rehabilitation movement classification using deep learning approach.

The rapid advancement of sensor technologies has sparked significant interest in intelligent physiotherapy monitoring systems, which hold immense potential for improving assessment quality. Despite this promise, existing sensor technologies for physiotherapy evaluation often need to be expanded in their versatility and robustness. This research introduces an innovative approach using spike train feature extraction to enhance patient progression tracking for physiotherapy assessment. developed a novel approach that reveals distinctive patterns for each physiotherapy exercise by implementing spike trains as the primary feature extraction method. The methodology used three datasets: UI-PRMD, K3Da, and a Self-Collected dataset, which were encoded into spike train formal representations, generating around 415 unique spike patterns. The study uses raster plot patterns as inputs for a sophisticated Deep Learning framework to assess pattern uniqueness. A key innovation was utilising spike occurrence frequency (firing rate) to differentiate movement correctness, where the derived mean error percentage (MPE) was used as a supportive metric complementing the classification process, and validated against DL evaluation metrics. The proposed framework demonstrated exceptional performance, achieving recognition rates of 99.44% (UI-PRMD), 98.21% (K3Da), and 100.00% (self-collected datasets) across various convolutional neural network architectures. Comprehensive evaluation metrics were used to validate the effectiveness of the rehabilitation movement assessment, including accuracy, precision, recall, and F1-score. Spike-train encoding combined with a tailored CNN is promising for physiotherapy movement recognition and correctness assessment, but clinical utility remains provisional and requires validation in patient populations like stroke and Parkinson’s disease.