A Deep Learning Approach for Classification of Physiotherapy Exercises Using Segmentation of Techniques

Abstract

Physiotherapy exercises are necessary to patients to restore their functional abilities in many cases as disabilities, injury, or basic with complementary approach as balneotherapy. Different type of exercised and different template sessions are used depending on the medical diagnostics. The evaluation of effectiveness of these exercises are important for patient’s rehabilitation process as time and level of recovery of locomotor skills. A dataset publicly available (Physical Therapy Exercises) is used for classification of session of repeated exercises that includes movement executed correct (C), fast execution (F) and low-amplitude execution (L). A novel approach is proposed by using segmentation of signal using deep learning neural network followed by a convolutional neural network for classification of sequence of the labeled classes L,C, F, and N (a new class introduced to label the noise of sensor of exercised or incorrect movement of the patient. The signal is extensively analyzed in order to made and corresponding labeling for analyzing using sliding window with a drive user selected length. The accuracy of classification is greater than 96% and sensitivity is greater than 95% but the results can be better if the labelling of N class is more restrictive and the effect of imbalanced dataset is reduced.

Keywords

• physiotherapy exercises
• segmentation techniques
• deep learning neural networks
• classification
• imbalanced dataset