Back to all papers

AI-Driven multi-view learning from CCTA for myocardial infarction diagnosis.

Authors

Gwizdala J,Salihu A,Senouf O,Meier D,Rotzinger D,Qanadli S,Muller O,Frossard P,Abbe E,Thanou D,Fournier S,Auberson D

Affiliations (5)

  • Institute of Mathematics, School of Computer and Communication Sciences, EPFL, Lausanne, Switzerland.
  • Signal Processing Laboratory, School of Engineering, EPFL, Lausanne, Switzerland.
  • Cardiology Department, University Hospital of Lausanne, University of Lausanne, Lausanne, Switzerland.
  • Radiology Department, University Hospital of Lausanne, Lausanne, Switzerland.
  • Cardiology Department, University Hospital of Lausanne, University of Lausanne, Lausanne, Switzerland. [email protected].

Abstract

Non-ST-elevation acute coronary syndrome (NSTE-ACS) remains a diagnostic challenge, as a proportion of patients do not present with obstructive coronary lesions. Coronary computed tomography angiography (CCTA) has emerged as a non-invasive tool for coronary assessment, and integrating artificial intelligence (AI) may enhance its diagnostic accuracy. This study evaluates a machine learning (ML) model using a learned fusion approach to identify culprit lesions in high-risk NSTE-ACS patients. This study is a sub-analysis of a prospective, multicenter trial including patients with high-risk NSTE-ACS who underwent CCTA, followed by ICA and fractional flow reserve (FFR) assessment in every intermediate stenosis. An ML framework was developed to analyze 2 orthogonal CCTA views of each coronary segment and classify them as culprit or non-culprit, with ICA +/- FFR as gold standards. The model was trained using 5-fold cross-validation and compared against 5 baseline methods, including conventional feature extraction and FFR-CT. Among 80 patients, 514 coronary segments were analyzed, with 63 (12.3%) labeled as culprit. The learned fusion model achieved a sensitivity of 0.55 ± 0.14, specificity of 0.93 ± 0.05, and F1-score of 0.53 ± 0.11. The AUC was 0.84 ± 0.06, matching the performance of FFR-CT (AUC of 0.82 ± 0.08). Our findings demonstrate that the learned fusion approach, based on combining two orthogonal views, achieved a performance level comparable to that of FFR-CT, as shown by the AUC of both techniques. These results confirm that AI-driven CCTA analysis could enhance clinical decision-making in high-risk NSTE-ACS patients, warranting further validation of this method in larger cohorts.

Topics

Journal Article

Ready to Sharpen Your Edge?

Join hundreds of your peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.