Non-invasive Technique to Automate the General Movements Assessment
G. Hewavitharana, I. Manawadu, R. Pytathma, D. P. M. H. Dharmananda, A. Wimalasena, S. Gunawardhana
This project features the first phase of designing a machine learning model to automate the General Movement Assessment using infant video library at Karapitiya Teaching Hospital.
Key Words: Cerebral Palsy, GMA, Automating GMA, Fidgety Movements, Writhing Movements,
Overview
General Movement Assessment (GMA) is a non-invasive way to detect neurological conditions that may contribute to Cerebral Palsy(CP) and other physical disorders. It can be carried out from childbirth to 3 months of age. The assessment is conducted by observing the child as they lay in a neutral environment recorded and examined by a well-trained panel of physicians. Therefore, coaching and preservation of assessment expertise are vitally important and it is a costly process to maintain the consistency of the of GMA. However, trained physicians are scarce, and scaling up the number of evaluations is hectic and may result in higher risks for losing accuracy.
On the other hand, automating the assessment increases the repeatability, availability, and in the long run can improve the accuracy of the diagnosis. There are standard Machine learning (ML) based models for face, object, and movement recognition. However, GMA requires the classification of movements to distinguish between normal and abnormal movements of an infant.
This work develops a machine learning model based on 2D image processing to distinguish between normal and abnormal movements and validate the model using a sample data set obtained from Paediatric Neuro Unit, Karapitiya Teaching Hospital. Due
to pandemic situation of the country, obtaining the ethical clearance to use the infant dataset got delayed and as the first step a machine learning model is developed to classify the different movements using a dataset generated using adults.
The Ethical clearance was obtained in April 2021 and the classification of infant movements are in progress.
Project Contacts: Dr. Subodha Gunawardena (subodha@eie.ruh.ac.lk)
Publications
- I. Manawadu, S. Gunawardena, N. Madhusanka, and D Rathnayake, “An Approach Towards Detecting Circular Fidgety Movements Using Machine Learning”, in Proceeding of 2nd IEEE INCET2021, May 2021.
- I. Manawadu, S. Gunawardena, L. Yasakethu, N. Madhusanka, and D Rathnayake, “Effect Of The Video Frame Rate On the Precision of Detecting Circular Fidgety Movements Using Machine Learning”, accepted for publication in Proceeding of ASIANCON 2021, Aug 2021