Fig. 1.

Overview of study design. A training set of segmented images from 1932 patients with multiple diseases from multiple centres were used to train four convolutional neural networks (CNNs). CNN segmentations were combined to measure left ventricular (LV) cavity volumes, systolic function and myocardial mass. Machine segmentations were compared to clinical segmentations on an independent dataset to measure precision. EDV end diastolic volume, ESV end systolic volume, EF ejection fraction, LVM LV mass, MV mitral valve, SAx short axis