Machine Learning and Similarity Network Approaches to Support Automatic Classification of Parkinson’s Diseases Using Accelerometer-based Gait Analysis

Abstract

Parkinson's Disease is a worldwide health problem, causing movement disorder and gait deficiencies. Automatic noninvasive techniques for Parkinson's disease diagnosis is appreciated by patients, clinicians and neuroscientists. Gait offers many advantages compared to other biometrics specifically when data is collected using wearable devices; data collection can be performed through inexpensive technologies, remotely, and continuously. In this study, a new set of gait features associated with Parkinson's Disease are introduced and extracted from accelerometer data. Then, we used a feature selection technique called maximum information gain minimum correlation (MIGMC). Using MIGMC, features are first reduced based on Information Gain method and then through Pearson correlation analysis and Tukey post-hoc multiple comparison test. The ability of several machine learning methods, including Support Vector Machine, Random Forest, AdaBoost, Bagging, and Naïve Bayes are investigated across different feature sets. Similarity Network analysis is also performed to validate our optimal feature set obtained using MIGMC technique. The effect of feature standardization is also investigated. Results indicates that standardization could improve all classifiers' performance. In addition, the feature set obtained using MIGMC provided the highest classification performance. It is shown that our results from Similarity Network analysis are consistent with our results from the classification task, emphasizing on the importance of choosing an optimal set of gait features to help objective assessment and automatic diagnosis of Parkinson's disease. Results illustrate that ensemble methods and specifically boosting classifiers had better performances than other classifiers. In summary, our preliminary results support the potential benefit of accelerometers as an objective tool for diagnostic purposes in PD.

Automatic radiomic feature extraction using deep learning for angiographic parametric imaging of intracranial aneurysms

“…The model’s performance on the testing cohort suggests that the network does not suffer from overfitting. There are published works in the context of medical machine learning applications that have training datasets (n=200),27 (n=30),28 (n=376)29 on a similar scale to ours. Additionally, our network task uses image data that are not as diverse as other computer vision applications.…”

Gait Analysis for Early Neurodegenerative Diseases Classification Through the Kinematic Theory of Rapid Human Movements

“…PD patients show slow automatic movements and slow balance. The symptoms are body rigidity with hypertonia, bradykinesia plus akinesia and lack of balance, especially when the disease is severe [12]. ALS shows an evolutionary muscular atrophy with decrease in strength, with phonation and chewing disorders [13].…”

Smart Computational Approaches with Advanced Feature Selection Algorithms for Optimizing the Classification of Mobility Data in Health Informatics

“…Improving the healthcare decision-making process using recent advancements in technology has been a trending domain in the past few decades. Technology has been used in the medical domain to develop automatic diagnosis, prognosis, and treatment evaluation process, decrease the number of patients' visits to the physicians' offices and reduce the cost of treatment [23]. The use of wearable movement monitoring devices together with machine learning (ML) techniques to improve diagnosis of patients with neurological diseases is one of the examples where movement parameters, including gait parameters, are used for classification of patients with neurological disorders from their healthy counterparts [23].…”

“…This, in turn, can help physicians and healthcare providers in making more informed decisions. Gait analysis has also been used for automatic recognition of gait due to aging [23]. Classification of the gait patterns of healthy young and healthy elderly individuals can have potential applications in age estimation [9].…”