Deep learning automatically distinguishes myocarditis patients from normal subjects based on MRI

• Published in: The international journal of cardiovascular imaging Vol. 40; no. 12; pp. 2617 - 2629
• Main Authors: Hatfaludi, Cosmin-Andrei, Roșca, Aurelian, Popescu, Andreea Bianca, Chitiboi, Teodora, Sharma, Puneet, Benedek, Theodora, Itu, Lucian Mihai
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.12.2024; Springer Nature B.V
• Subjects: Accuracy; Adult; Algorithms; Arrhythmia; Automation; Cardiac arrhythmia; Cardiac Imaging; Cardiology; Case-Control Studies; Congestive heart failure; Consent; Datasets; Deep Learning; Female; Gadolinium; Heart diseases; Heart failure; Humans; Image Interpretation, Computer-Assisted; Imaging; Machine learning; Magnetic Resonance Imaging; Magnetic Resonance Imaging, Cine; Male; Medicine; Medicine & Public Health; Middle Aged; Myocarditis; Myocarditis - diagnostic imaging; Myocarditis - physiopathology; Myocardium - pathology; Neural networks; Original Paper; Patients; Predictive Value of Tests; Radiology; Reproducibility of Results; Retrospective Studies; Young Adult; Myocarditis; Deep neural networks; Medical imaging; Few-shot learning; CMRI
• ISSN: 1875-8312; 1569-5794; 1875-8312; 1573-0743
• DOI: 10.1007/s10554-024-03284-8

Myocarditis, characterized by inflammation of the myocardial tissue, presents substantial risks to cardiovascular functionality, potentially precipitating critical outcomes including heart failure and arrhythmias. This investigation primarily aims to identify the optimal cardiovascular magnetic resonance imaging (CMRI) views for distinguishing between normal and myocarditis cases, using deep learning (DL) methodologies. Analyzing CMRI data from a cohort of 269 individuals, with 231 confirmed myocarditis cases and 38 as control participants, we implemented an innovative DL framework to facilitate the automated detection of myocarditis. Our approach was divided into single-frame and multi-frame analyses to evaluate different views and types of acquisitions for optimal diagnostic accuracy. The results demonstrated a weighted accuracy of 96.9%, with the highest accuracy achieved using the late gadolinium enhancement (LGE) 2-chamber view, underscoring the potential of DL in distinguishing myocarditis from normal cases on CMRI data.