Deep Learning-Based Classification of Neurodegenerative Diseases Using Gait Dataset: A Comparative Study

Neuro-degenerative diseases pose a significant health concern for human society, especially among the elderly population. Given their high prevalence and the limited availability of clinical expertise and services, there is an urgent requirement for the integration of artificial intelligence (AI) systems to support healthcare professionals in addressing this issue. This paper presents a comparative analysis of seven deep learning architectures for neuro-degenerative disease classification by using a gait dynamics dataset. The models used in the analysis include LSTM, GRU, InceptionTime, ResNet, FCN, TST, and PatchTST. The models are extensively evaluated for different classification tasks. The findings of the study suggest that deep learning techniques have the potential to diagnose and classify different neurodegenerative diseases effectively. It can be inferred that ResNet exhibits superior performance compared to other models in tasks involving the classification of healthy controls (HC) and class of individuals with all neuro-degenerative diseases (NDDs), and classification of healthy controls (HC) from individuals with Parkinson's disease (PD). TST demonstrates the highest level of performance among all models in tasks involving distinguishing between Amyotrophic Lateral Sclerosis (ALS) and healthy controls (HC), as well as Huntington's disease (HD) and healthy controls (HC).