KNN-BASED EMG 2 DOF SYSTEM FOR POST-STROKE PATIENT THERAPY: FINGER MOVEMENT CLASSIFICATION IN HAND REHABILITATION

• Main Author: Khoirurrahma, Amelia
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/79486
• Institution: Institut Teknologi Bandung

Routine limb therapy for post-stroke patients to restore coordination and muscle strength in limbs that have weakened after being affected by the disease is important. Stepping exercises and mirror neuron therapy are two limb rehabilitation techniques that might assist patients in strengthening their muscles. However, usually therapy uses both methods separately. It is necessary to develop a therapy method that can combine the two to increase the success of therapy for post-stroke patients. This method needs a sensor system is needed to read and classify healthy hand movements effectively and the signal received by the driving system to move the affected part of the hand as part of the therapy method. This Final Project research develop a sensor system for reading different variations of healthy hand finger movements using a 2 Degree of Freedom (DoF) system based on two channels electromyography (EMG) and classifying them in real-time using muscle sensor v3. The classification algorithm used in this research is KNN (K- Nearest Neighbor). Tests involves three muscle configurations for two-channel electrode placement and several statistical features as classification parameters. The results obtained show that the best electrode placement configuration is located on the extensor indicis, extensor pollicis brevis, extensor pollicis longus, and abductor pollicis longus muscle area for the first channel and on the extensor digiti minimi, extensor digitorum, and extensor calpi ulnaris muscle area for the second channel. The best combination of statistical features to use as classification parameters is the mean and median. Based on these parameters, the classification results for seven variations of finger movements show accuracy values reaching 0.705, precision 0.716, sensitivity 0.706, and specificity 0.951. The accuracy value can increase to 0.77 for classification of 4 movements and 0.98 for classification of 2 finger movements.