Statistical Analysis of Gait Data to Assist Clinical Decision Making

Abstract

Gait analysis is used for non-automated and automated diagnosis of various neuromuskuloskeletal abnormalities. Automated systems are important in assisting physicians for diagnosis of various diseases. This study presents preliminary steps of designing a clinical decision support system for semi-automated diagnosis of knee illnesses by using temporal gait data. This study compares the gait of Ill patients with 110 age-matched normal subjects. Different feature reduction techniques, (FFT, averaging and PCA) are compared by the Mahalanobis Distance criterion and by performances of well known classifiers. The feature selection criteria used reveals that the gait measurements for different parts of the body such as knee or hip to be more effective for detection of the illnesses. Then, a set of classifiers is tested by a ten-fold cross validation approach on all datasets. It is observed that average based datasets performed better than FFT applied ones for almost all classifiers while PCA applied dataset performed better for linear classifiers. In general, nonlinear classifiers performed quite well (best error rate is about 0.035) and better than the linear ones.