Fig. 6

Example segmentations by machine learning algorithm. Top row: a pair of diastole images from the scan:rescan dataset that has been segmented by the automated algorithm. Note that the LV metrics are not exactly the same due to intrinsic variability in how slices are prescribed. Bottom left: example of an error (1 in 479 error rate) where laminar thrombus had been mis-identified as myocardium since this had not been ‘seen’ in the training data before. Bottom right: a mis-segmentation due to a pericardial effusion