One model to segment the heart across CT and MRI

Abdul Qayyum, Moona Mazher and Steven Niederer presented CardioSeqM, a single model that segments the left atrium, the whole heart and myocardial pathology, in the MICCAI challenge proceedings on real-world medical images.

CardioSeqM first learns from unlabelled cardiac scans, through self-supervised pretraining, and then reads each 3D volume as a sequence, so that it sees the structure of the whole heart as well as its local detail. The same model segments the left atrium in late gadolinium enhancement MRI, the whole heart in CT and in MRI, and myocardial pathology.

Published in the proceedings of the MICCAI Challenge on Comprehensive Analysis and Computing of Real-World Medical Images, Lecture Notes in Computer Science, doi.org/10.1007/978-3-032-16271-7_7.