Gait analysis and Machine learning algorithms to distinguish between Parkinson’s disease, amyotrophic lateral sclerosis, and Huntington’s disease

Abstract

Parkinson’s disease is a neurological disorder that is both common and complex. It affects the nervous system, resulting in tremors, sluggish movements, and imbalance. It is a chronic and progressive condition that is regarded as the most prevalent neurodegenerative disease following Alzheimer’s disease. Although timely diagnosis enables the provision of essential care, Parkinson’s disease identification can be difficult due to the presence of numerous neurological disorders. Movement disorders are early symptoms of neurological diseases, and motion dynamics is an effective method for diagnosing them. This method evaluates and diagnoses the disease by analyzing the gait patterns of patients. This study specifically demonstrated this method to distinguish between Parkinson’s disease and other neurological disorders. After being obtained from the Physiont database, the proposed method preprocessed signals. The database contained 15 signals from individuals with Parkinson’s disease and 49 signals from healthy individuals and those with other neurological conditions. The wavelet transform filter bank with default coefficients from MATLAB software was employed to denoise and enhance the acquired signals in order to achieve superior results. The fourth variant of Dubichs, which was appropriate for biomedical signals, implemented a discrete wavelet transform with eight decomposition levels. Subsequently, a set of statistical, temporal, frequency, and nonlinear characteristics of signal was extracted and prioritized according to energy. Once the optimal features were selected from these extractions, linear support vector machines (SVM) and nonlinear models (nearest neighbor (KNN) and multilayer perceptrons (MLPs) were employed to classify signal information. The proposed method’s validity was evaluated through confusion matrix analysis, as well as calculations of specificity, sensitivity, and accuracy. The results suggest that the SVM classifier had a substantial discriminative capacity for distinguishing between Parkinson’s and non-Parkinson’s patients. The support vector machine was the most accurate classifier, achieving an accuracy rate of 98.4% in its ability to distinguish between individuals with Parkinson’s disease and those without. Multilayer perceptrons (MLPs), linear support vector machines (SVM), and nonlinear models (KNN) were used to classify the signal information after the optimal features from these extractions were determined. The proposed method was evaluated by confusion matrix analysis, specificity, sensitivity, and accuracy. The results demonstrate that the SVM classifier was highly discriminative in differentiating patients with and without Parkinson's disease, with an overall accuracy of 98.4% when differentiating individuals with and without Parkinson's disease, making the support vector machine the most accurate classifier.