Deep learning promises to revolutionise cardiac imaging with more accurate and more reproducible interpretation delivered almost instantaneously. But transferring promise into clinical delivery is a considerable task and we lag behind other fields.
The growing demand for cardiac imaging due to an ageing population with growing disease burden is not matched by an increased supply of clinicians trained in image interpretation. This is exacerbated in congenital heart disease due to increasing survivorship, disease complexity and paucity of available expertise. Deep learning has revolutionised the field of image analysis and there are now systems capable of automating cardiac image segmentation. Eliminating tedious tasks such as measuring structures on images can reduce burn-out and refocus clinicians on higher-level functions, the doctor–patient relationship and providing more empathic care.
But deep learning promises much more: it has the potential to improve on expert human image interpretation, and this is being realised with reports of better-than-human performance for breast cancer detection.1 We, as clinical experts, are not as good as we think we are at image interpretation—the minimal detectable change when ejection fraction is measured by an expert on cardiac MRI is 8.7%, mostly due to poor repeatability.2 Deep learning has potential for more accurate and more precise analysis where confidence in reported results can be quantified. Benefits could cascade through healthcare systems, with fewer repeated tests, less follow-up, and ultimately better clinical outcomes.
Cheap hardware and easy-to-use open-source software have made deep …
[1]
H. Baumgartner,et al.
Prediction of prognosis in patients with tetralogy of Fallot based on deep learning imaging analysis
,
2020,
Heart.
[2]
David S. Melnick,et al.
International evaluation of an AI system for breast cancer screening
,
2020,
Nature.
[3]
E. Topol,et al.
A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis.
,
2019,
The Lancet. Digital health.
[4]
D. Rueckert,et al.
A Multicenter, Scan-Rescan, Human and Machine Learning CMR Study to Test Generalizability and Precision in Imaging Biomarker Analysis.
,
2019,
Circulation. Cardiovascular imaging.
[5]
K. Gauvreau,et al.
Contemporary predictors of death and sustained ventricular tachycardia in patients with repaired tetralogy of Fallot enrolled in the INDICATOR cohort
,
2013,
Heart.
[6]
L. Harris,et al.
Implantable Cardioverter-Defibrillators in Tetralogy of Fallot
,
2008,
Circulation.
[7]
R. Berger,et al.
Are Implantable Cardioverter Defibrillator Shocks a Surrogate for Sudden Cardiac Death in Patients With Nonischemic Cardiomyopathy?
,
2006,
Circulation.
[8]
R. Franklin,et al.
Prognostic significance of ventricular arrhythmia after repair of tetralogy of Fallot: a 12-year prospective study.
,
1994,
Journal of the American College of Cardiology.