An Explainable AI-based Motion Detection approach for MR images without requirement of motion annotated ground truth data

Motivation: Motion is one of the leading causes of artifacts in MR images. Such images are often rendered non-diagnostic, and results in patient recall. Goal(s): Objective is to detect motion in MRI scans and provide information on extent of motion so that MRI technologists can decide on whether to re-scan the subject. Approach: The proposed AI-based method detects motion in MR images and provides a motion severity score, while not relying on real world annotated motion data. Results: The proposed method was tested on a set of real-world MRI data with and without motion and provided an accuracy of 89.6%. Impact: Reliable motion alert for MRI scans shall enable technologists to re-scan the subjects while in the scanning room. This will help in reducing the patient recalls due to motion artifact and hence reduce burden on the healthcare system.