Deep Learning Based Classification and Segmentation for Cardiac Magnetic Resonance Imaging with Respiratory Motion Artifacts

Abstract

Cardiac Magnetic Resonance (CMR) is key in the evaluation of heart anatomy and function, and the diagnosis of multiple diseases. However, it requires extensive manual analysis by medical specialists, which slows the diagnostic process, and creates an additional burden for professionals. Different computational techniques have been proposed to automate and accelerate the segmentation of different heart structures within the images, and with the rise of Deep Learning (DL) techniques in the last decade, the performance has improved significantly. Nevertheless, there are still some limitations to be addressed in the automatic processing of CMR, being the respiratory motion artifacts the focus of this paper. This paper presents a DL-based approach for the task of image quality classification and segmentation of heart structures in the context of the CMRxMotion public challenge.