Hybrid Feature Subset Selection for the Quantitative Assessment of Skills of Stroke Patients in Activity of Daily Living Tasks

Abstract

Stroke patients have a decreased ability in performing activity of daily living (ADL) tasks such as in `drinking a glass of water', `turning a key', `picking up a spoon', `lifting a bag', `reaching a bottle' and `lifting and carrying a bottle'. These tasks can be quantified by measuring forces and torques exerted on the objects. However, the resulting force and torque time series represent information at a very low level of abstraction and don't inform clinicians what really distinguishes patients from normal controls in performing these tasks. We conduct an extensive quantitative analysis of these tasks and derive interesting features from the time signals that characterize the differences in behavior between patients and normal controls. We show that `drinking a glass' and `turning a key' are the most discriminative tasks; furthermore we show that the ability or disability to synchronize the thumb and the middle finger is one of the most important features