Automatic detection of motion artifacts in MR images using CNNS

Abstract

Considerable practical interest exists in being able to automatically determine whether a recorded magnetic resonance image is affected by motion artifacts caused by patient movements during scanning. Existing approaches usually rely on the use of navigators or external sensors to detect and track patient motion during image acquisition. In this work, we present an algorithm based on convolutional neural networks that enables fully automated detection of motion artifacts in MR scans without special hardware requirements. The approach is data driven and uses the magnitude of MR images in the spatial domain as input. We evaluate the performance of our algorithm on both synthetic and real data and observe adequate performance in terms of accuracy and generalization to different types of data. Our proposed approach could potentially be used in clinical practice to tag an MR image as motion-free or motion-corrupted immediately after a scan is finished. This process would facilitate the acquisition of high-quality MR images that are often indispensable for accurate medical diagnosis.