Weakly supervised classification of aortic valve malformations using unlabeled cardiac MRI sequences

Abstract

Biomedical repositories such as the UK Biobank provide increasing access to prospectively collected cardiac imaging, however these data are unlabeled, which creates barriers to their use in supervised machine learning. We develop a weakly supervised deep learning model for classification of aortic valve malformations using up to 4,000 unlabeled cardiac MRI sequences. Instead of requiring highly curated training data, weak supervision relies on noisy heuristics defined by domain experts to programmatically generate large-scale, imperfect training labels. For aortic valve classification, models trained with imperfect labels substantially outperform a supervised model trained on hand-labeled MRIs. In an orthogonal validation experiment using health outcomes data, our model identifies individuals with a 1.8-fold increase in risk of a major adverse cardiac event. This work formalizes a deep learning baseline for aortic valve classification and outlines a general strategy for using weak supervision to train machine learning models using unlabeled medical images at scale.

Snorkel: rapid training data creation with weak supervision

“…Crowdsourcing (Crowd) We trained a model to perform sentiment analysis using crowdsourced annotations from the weather sentiment task from Crowdflower. 17 In this task, contributors were asked to grade the sentiment of oftenambiguous tweets relating to the weather, choosing between five categories of sentiment. Twenty contributors graded each tweet, but due to the difficulty of the task and lack of crowdworker filtering, there were many conflicts in worker labels.…”

“…For example, in the EHR experiment, where we had access to a large unlabeled corpus, we were able to achieve significant gains (8.1 F1 score points) in going from 100 to 50 thousand documents. Further empirical validation of these strong unlabeled scaling results can be found in follow-up work using Snorkel in a range of application domains, including aortic valve classification in MRI videos [17], industrial-scale content classification at Google [4], fine-grained named entity recognition [45], radiology image triage [26], and others. Based on both this empirical validation, and feedback from Snorkel users in practice, we see this ability to leverage available unlabeled data without any additional user labeling effort as a significant advantage of the proposed weak supervision approach.…”

Cardiac imaging of aortic valve area from 26,142 UK Biobank participants reveal novel genetic associations and shared genetic comorbidity with multiple disease phenotypes

“…the cardiac cycle (Figure 1). A more detailed description of the initial protocol, including supporting videos, is presented by Fries et al 16 .…”

“…While diagnosis of aortic valve stenosis may be derived from a combination of findings from physical exam, auscultation, echocardiography or functional data obtained during cardiac catheterization or surgery, recent findings support the clinical utility of software-based methods for magnetic resonance imaging (MRI) analysis 13 . Novel magnetic resonance imaging (MRI)-based techniques to automatically distinguish between bicuspid and normal (tricuspid) aortic valve have potential to translate into clinical applications 14,15 , and to facilitate biomedical research through the use of large biobanks 16 . Here, we use planimetric measures of the aortic valve functional area obtained through an automated approach for cardiac MRI data 16 as a proxy for valvular function, to conduct a genome-wide association study on 26,142 individuals of European ancestry from the UK Biobank.…”

“…Novel magnetic resonance imaging (MRI)-based techniques to automatically distinguish between bicuspid and normal (tricuspid) aortic valve have potential to translate into clinical applications 14,15 , and to facilitate biomedical research through the use of large biobanks 16 . Here, we use planimetric measures of the aortic valve functional area obtained through an automated approach for cardiac MRI data 16 as a proxy for valvular function, to conduct a genome-wide association study on 26,142 individuals of European ancestry from the UK Biobank. We compute polygenic scores and perform a phenome-wide association study to search for potential links between the genetic basis of aortic valve size and other health-related phenotypes.…”