Scalogram-Based Gait Abnormalities Classification Using Deep Convolutional Networks for Neurological and Non-Neurological Disorders

In the present day, there is a steep increase in cases of neurological and non-neurological diseases that may affect a person’s normal gait. This study proposes a methodology for classifying gait abnormalities using convolutional neural networks based on scalogram analysis of electromyography and foot insole data.

This study utilizes scalograms for classifying electromyography and foot-insole data, offering robust handling of high-noise data with sudden transitions. The electromyography data is sourced from patients with hemiplegia, rheumatoid arthritis, prolapsed intervertebral disc, and osteoarthritis. Foot insole data from PhysioNet includes subjects with Parkinson’s disease, Huntington’s disease, and amyotrophic lateral sclerosis. Classification tasks are performed using convolutional neural network architecture, with performance assessed across three optimizers—adam, stochastic gradient descent, and rmsprop—and employing both max and average pooling layers for enhanced model optimization.

The results indicate promising classification performance, with an accuracy of 96.75% achieved using the adam optimizer with max pooling layer for overall classification. For the four-class classification task, 96.62% and 96.96% accuracy were attained using adam and rmsprop optimizers with max-pooling layers, respectively.

The study demonstrates that scalogram-based analysis, coupled with CNN classification, provides a robust framework for accurate and reliable diagnosis of gait abnormalities. The methodology shows significant promise in differentiating between different types of gait disorders, offering potential applications in clinical settings for reliable and accurate diagnosis of gait abnormalities.