Towards a Deep Learning based Approach for an Infant Medical Analysis-A Review

Abstract

Motion analysis is one of the known ways used to establish whether an infant is normal or abnormal. Studies have indicated that there is a clear difference in terms of speed or even time taken to respond to stimuli. General movements (GMs) are spontaneous movements of infants that involves the entire body differing in speed, amplitude and sequence. The assessment of GMs has helped in identifying infants that are at risk of neurological disorders. GMs assessment is based on videos recorded by parents or caregivers which are then rated by a clinicians or trained professionals. The General Assessment Tool has worked well, however, it is time consuming and very expensive. Several techniques have been proposed to automate the General Movement assessment tool which include markerbased techniques and markerless techniques. In our review we have systematically discussed the design features and technologies involved in both of them and identified both the strength and weakness. Thereafter, we explain the reasons for their limited practical performance. We conclude by proposing a deep learning approach that can be used to possibly address the issues raised in the existing techniques.