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

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Non-ST-elevation acute coronary syndrome (NSTE-ACS) presents diagnostic challenges, with some patients lacking obstructive coronary artery disease.
• Coronary computed tomography angiography (CCTA) is a non-invasive imaging tool for coronary assessment.
• Artificial intelligence (AI) integration may improve CCTA's diagnostic accuracy.

Purpose of the Study:

• To evaluate a machine learning (ML) model using a learned fusion approach for identifying culprit lesions in high-risk NSTE-ACS patients.
• To assess the diagnostic performance of the ML model compared to established methods.

Main Methods:

• A sub-analysis of a prospective, multicenter trial involving high-risk NSTE-ACS patients who underwent CCTA.
• Development of an ML framework to analyze orthogonal CCTA views, classifying coronary segments as culprit or non-culprit using invasive coronary angiography (ICA) +/- fractional flow reserve (FFR) as the gold standard.
• Model training via 5-fold cross-validation and comparison against conventional feature extraction and FFR-CT.

Main Results:

• Analysis of 514 coronary segments from 80 patients, with 63 (12.3%) identified as culprit lesions.
• The learned fusion ML model achieved a sensitivity of 0.55 ± 0.14, specificity of 0.93 ± 0.05, and F1-score of 0.53 ± 0.11.
• The area under the curve (AUC) for the ML model was 0.84 ± 0.06, comparable to FFR-CT (AUC 0.82 ± 0.08).

Conclusions:

• The AI-driven learned fusion approach for CCTA analysis demonstrates performance comparable to FFR-CT in identifying culprit lesions.
• AI-enhanced CCTA analysis holds promise for improving clinical decision-making in high-risk NSTE-ACS patients.
• Further validation in larger patient cohorts is warranted to confirm the utility of this AI method.