Learning-Based Quality Control for Cardiac MR Images
G. Tarroni, O. Oktay, W. Bai, A. Schuh, H. Suzuki, J. Passerat-Palmbach, A. de Marvao, D. P. O'Regan, S. Cook, B. Glocker, P. M. Matthews, D. Rueckert
IEEE Trans Med Imaging., vol. 37, issue 11, November 2018
The effectiveness of a cardiovascular magnetic resonance (CMR) scan depends on the ability of the operator to correctly tune the acquisition parameters to the subject being scanned and on the potential occurrence of imaging artefacts such as cardiac and respiratory motion. In the clinical practice, a quality control step is performed by visual assessment of the acquired images. However, this procedure is strongly operator-dependent, cumbersome and sometimes incompatible with the time constraints in clinical settings and large-scale studies. On the other hand, failure to correctly identify corrupted or unusable images could affect the results of automated analysis performed on the acquired data with undesirable effects.
In this paper we present a fast, fully-automated, learning-based quality control pipeline for CMR images, specifically for short-axis image stacks. The pipeline performs three important quality checks: 1) heart coverage estimation, 2) inter-slice motion detection and 3) image contrast estimation in the cardiac region. The technique has been tested on thousands of cases extracted from the UK Biobank and the UK Digital Heart Project, and has been found accurate and robust when compared against manual annotations and visual inspections performed by expert interpreters. As a consequence, it could be potentially deployed both as almost real-time tool at acquisition site and as pre-processing step before conventional image analysis, ensuring the reliability of the obtained results.