A Framework for Automated Cardiovascular Magnetic Resonance Image Quality Scoring based on EuroCMR Registry Criteria

Cardiovascular magnetic resonance (CMR) imaging is a radiation-free modality widely used for functional and structural evaluation of the cardiovascular system. Achieving an accurate diagnosis requires having good-quality images. Subjective CMR image quality assessment is a tedious, time-consuming and costly process. This paper presents an automated scoring framework for CMR image quality assessment that uses deep learning models to evaluate left ventricular coverage and CMR imaging artefacts. The quality scoring in the proposed framework is an attempt to automate some of the subjective quality control criteria of the EuroCMR registry for the short-axis cine steady-state free precession (SSFP) CMR images. The scores given by a radiologist and a cardiologist with experience in CMR imaging for the images of 50 subjects from the UK Biobank were used to validate the proposed framework. The Pearson correlation coefficient (PCC) and the Spearman rank-order correlation coefficient (SRCC) calculated for the experts' quality scores versus ones obtained from the proposed framework are 0.908 and 0.806 on average. The results show that the quality scoring by the proposed framework is highly correlated with the experts' opinions. The proposed framework can be used for post-imaging quality assessment of short-axis cine SSFP CMR images and quality control of large population studies such as the UK Biobank.