Weakly supervised classification of rare aortic valve malformations using unlabeled cardiac MRI sequences
Jason A. Fries, Paroma Varma, Vincent S. Chen, Ke Xiao, Heliodoro Tejeda, Priyanka Saha, Jared Dunnmon, Henry Chubb, Shiraz Maskatia, Madalina Fiterau, Scott Delp, Euan Ashley, Christopher Re, James R. Priest
Received Date: 9th August 2018
Population-scale biomedical repositories such as the UK Biobank provide unprecedented access to prospectively collected cardiac imaging data, however the majority of these data are unlabeled, creating barriers to their use in supervised machine learning. We developed a weakly supervised deep learning model for Bicuspid Aortic Valve (BAV) classification using up to 4,000 unlabeled cardiac MRI sequences. Instead of requiring curated, hand-labeled training data, weak supervision relies on noisy heuristics defined by domain experts to programmatically generate large-scale, imperfect training labels. For BAV classification, training models using these imperfect labels substantially outperformed a traditional supervised model trained on hand-labeled MRIs. In a validation experiment using long-term outcome data from the UK Biobank, our classification model identified a subset of individuals with a 1.8-fold increase in risk of a major adverse cardiac event. This work formalizes the first deep learning baseline for aortic valve classification and outlines a general strategy for using weak supervision to train machine learning models using large collections of unlabeled medical images.
Read in full at bioRxiv.
This is an abstract of a preprint hosted on an independent third party site. It has not been peer reviewed but is currently under consideration at Nature Communications.