Classification of Gait Abnormalities Using Transfer Learning with EMG Scalogram Features

Abstract

Gait abnormalities can significantly impact the mobility and quality of life of individuals, thus making its early diagnosis crucial for proper treatment planning and rehabilitation. In this study scalograms generated from EMG signals of two important gait abnormalities, rheumatoid arthritis, and prolapsed intervertebral disc are classified using transfer learning. Scalogram has an advantage when dealing with high-noise data with abrupt transitions making it an excellent choice for classifying movement patterns. When classified using only CNN an accuracy of 91.1%, precision of 92.8%, recall of 92.9%, AUC of 0.98, and a PRC of 0.97 were obtained. For transfer learning, VGG16, VGG19, ResNet50, Inceptionv3, InceptionResNet, MobileNet and MobileNetv2, and DenseNet Large were incorporated along with previous CNN. DenseNet Large achieved highest accuracy of 97.5% along with 96.2% precision, 96.2% recall, an AUC of 0.99 and a PRC of 0.99. The use of transfer learning provided a significant boost to performance of the model. The proposed method of using scalograms with transfer learning can be used to accurately diagnose gait abnormalities and allow medical professionals to design treatment and rehabilitation plan. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2023.