Identification and Analysis of Neurodegenerative Diseases with Twin Layered CNN Using Gait Dynamics
International Journal of Intelligent Engineering and Systems
This study identifies the gait dynamics of Neurodegenerative disease (NDD) patients using a deep convolution neural network (CNN) approach. Human gait examination perceives individuals from the manner in which they walk, which gives clues in identifying specific disorders. Precise identification of the specific NDD assists the doctors to begin early treatment strategies. The physionet public database containing the gait dynamics of Parkinson's, Amyotrophic Lateral Sclerosis, Huntington's
... s, and the healthy subjects is used in this work. The gait parameters such as left stride intervals, right stride intervals, left swing intervals, right swing intervals, left stance intervals and right stance intervals of each disease group are utilized. Deep Twin layered CNN (TLCNN) is applied as a feature extractor, which automatically extracts deep features from the input gait signals. Further statistical features are also measured for these inputs by implementing Fast Walsh Hadamard Transform method (FWHT). The tuned neighbourhood component analysis (TNCA) is utilized as a feature selection algorithm to select prime features from the extracted deep feature vector. The classification of each disease group is made with the selected features by the Random forest and Multi SVM machine learning approach, which gives the accuracy of 99.89%, and 97.06% in classification of NDD and healthy person. Thus, the proposed TLCNN_FWHT_TNCA_RF gait classification approach based on CNN and machine learning is accurate and reliable to classify the particular neurodegenerative disorder.