Applying machine learning to gait analysis data for disease identification.

A machine-learning framework to identify the specific disease afflicting certain patients diagnosed with Neurological and Neuromuscular Diseases (NND) or Juvenile Idiopathic Arthritis (JIA) using only gait analysis data is presented. Classifying such data into disease types consumes valuable clinical time that may be better spent. Effective classification also facilitates its future retrieval. To prove the feasibility of the approach, we applied it to the simpler case of identifying the disease class of patients with a view to extending the method to specific diseases in future work. Standard clinical gait information of healthy individuals, and NND/JIA patients was sourced from hospitals participating in MD-PAEDIGREE. To classify the data into one of the three categories: healthy, NND, and JIA, certain parameters were carefully selected from the signals and used to train Random Forest (RF), boosting, Multilayer Perceptron (MLP), and Support Vector Machine (SVM) classifiers. Cross-validation was used to test the effectiveness of our approach and it yields a classification accuracy of 100% for RF, SVM, and MLP classifiers and 96.4% for boosting. Training and testing for all the classifiers took mere milliseconds, providing opportunities for real-time applications. To extend the method to the identification of specific illnesses, more discerning features from the gait data are currently being investigated. Moreover, a larger dataset is being gathered. Finally, we are attempting to reduce the number of features used for classification in order to further decrease computation time and algorithm complexity.