Clinical InvestigationMachine Learning Detection of Aortic StenosisAutomated Detection of Aortic Stenosis Using Machine Learning
Section snippets
Echocardiograms
This work was approved by the Tufts Medical Center institutional review board. The echocardiograms originate from TTE examinations performed between 2011 and 2020 at a high-volume tertiary care center (Tufts Medical Center). The echocardiograms were acquired as part of routine clinical care. The CardioVascular Imaging Center is Intersocietal Accreditation Commission accredited and is equipped with ultrasound units from major vendors (Philips, Toshiba, and Siemens). By using standardized Digital
Results
The clinical characteristics of the patients included in this study are shown in Table 3. The primary labeled cohort included 577 patients. The median age was 74 years (interquartile range, 63-82 years). Forty-three percent of the patients were women. Eighty-six percent of the study population was Caucasian. The hemodynamic parameters of the echocardiograms are shown in Table 4. The median AoV peak velocity was 2.89 m/sec (interquartile range, 2.29-3.67 m/sec), the median peak gradient was
Discussion
Novel approaches to AS case identification are needed to improve treatment rates for this condition. Here we develop methods for fully automated detection of AS from limited TTE data sets. We show that automated detection of AS is possible using modern deep learning classifiers and that these networks are generalizable across different data sets. These tools can broadly characterize the presence or absence of AS and the severity of disease and are well suited for identifying patients who should
Conclusion
ML approaches optimized for echocardiography can successfully identify AS using limited 2D data sets. These methods lay the groundwork for fully automated screening for this disease and future study of interventions to improve outcomes.
References (29)
- et al.
Evaluating medical therapy for calcific aortic stenosis: JACC state-of-the-art review
J Am Coll Cardiol
(2021) - et al.
Trends in utilization of aortic valve replacement for severe aortic stenosis
J Am Coll Cardiol
(2022) - et al.
Recommendations on the echocardiographic assessment of aortic valve stenosis: a focused update from the European Association of cardiovascular imaging and the American Society of echocardiography
J Am Soc Echocardiogr
(2017) - et al.
Automated interpretation of systolic and diastolic function on the echocardiogram: a multicohort study
Lancet Digit Heal
(2022) - et al.
A machine-learning framework to identify distinct phenotypes of aortic stenosis severity
JACC Cardiovasc Imaging
(2021) - et al.
Enhanced diagnosis of severe aortic stenosis using artificial intelligence: a proof-of-concept study of 530,871 echocardiograms
JACC Cardiovasc Imaging
(2020) - et al.
Identifying aortic stenosis with a single parasternal long-Axis video using deep learning
J Am Soc Echocardiogr
(2023) - et al.
Guidelines for performing a comprehensive transthoracic echocardiographic examination in adults: recommendations from the American society of echocardiography
J Am Soc Echocardiogr
(2019) - et al.
Multi reader assessment of accuracy and interobserver variability in aortic stenosis by echocardiography
Hear Lung Circ
(2019) - et al.
Proposed requirements for cardiovascular imaging-related machine learning evaluation (PRIME): a checklist
JACC Cardiovasc Imaging
(2020)
Aortic valve stenosis treatment disparities in the underserved: JACC council perspectives
J Am Coll Cardiol
Trends in transcatheter and surgical aortic valve replacement in the United States, 2008-2018
Am Heart J
Provider-level variability in the treatment of patients with severe symptomatic aortic valve stenosis
J Am Coll Cardiol
Global, regional, and national burden of calcific aortic valve and degenerative mitral valve diseases, 1990-2017
Circulation
Cited by (8)
Echoing Errors: The Problem of Uncurated “Big Data” in Echocardiography
2023, Journal of the American Society of EchocardiographyDetecting Heart Disease from Multi-View Ultrasound Images via Supervised Attention Multiple Instance Learning
2023, Proceedings of Machine Learning Research
This work is supported by National Center for Advancing Translational Sciences grant UL1TR002544. Dr Wessler is supported by National Institutes of Health grant K23AG055667.
This work was supported by the National Institutes of Health Tufts CTSI (NIH CTSA UL1TR002544). Dr Wessler received funding from the National Institutes of Health (K23 AG055667).
Dr Wessler has done consulting work with iCardio.ai and US2.ai unrelated to the present work and is a cofounder of CVAI Solutions. CVAI Solutions created the software for the deidentification procedures but currently has no related commercial pursuits.