A time-induced interval type-2 fuzzy approach towards functional recovery of stroke patients

Upper limb movement abnormalities are frequent as a result of stroke or brain injury. Following this, the patient might recover gradually with therapeutic inputs and ongoing supervision of rehabilitative exercises. Compensatory work during such exercises involves completing that task in the simplest method possible, which results in incorrect outputs. The current work begins with extracting pertinent features, such as the angle and distance between the relevant body joints, from kinetic data. We have used a novel time-induced interval type-2 fuzzy set on the feature vector by adding a time axis in addition to the two axes taken into consideration in the conventional one, to model the membership values of the features concerning the time frame. The fuzzified feature values are then fed to a recurrent neural network to detect the compensation generated during the exercises. The proposed framework's 98.45% accuracy has shown its wide applicability for automated training quality improvement.