An Approach Towards Detecting Circular Fidgety Movements Using Machine Learning

Cerebral Palsy (CP) is a neurological condition that is caused by brain damage and it is considered as the most common movement disability of childhood. General Movements Assessment (GMA) is a non-invasive and cost-effective way of identifying CP. At the age of six to nine weeks, characteristics of general movements of a normal infant vary from Writhing Movement (WM) type to Fidgety Movement (FM) type. One of the available methods of diagnosing CP is by analyzing fidgety circular movement patterns. The aim of the present study is to introduce a novel machine learning model to distinguish circular movements out of the rest of the movements using a Kinect sensor. A movement dataset was generated by using adults and a recurrent neural network was used for the classification. In accordance to the circular movements identification, this model had a sensitivity (91.3%), specificity (85.7%) and precision (88.45%). This model can be used with infant data to distinguish circular fidgety movements from rest of the movements after appropriate training.