Identifying Pathogenesis of Acute Coronary Syndromes using Sequence Contrastive Learning in Coronary Angiography

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Acute coronary syndrome (ACS) diagnosis relies on identifying underlying mechanisms like plaque rupture (PR) and plaque erosion (PE).
• Distinguishing PR from PE is crucial for tailoring treatment, with PR typically managed by stenting and PE potentially managed conservatively.
• Current methods often require invasive intracoronary imaging, prompting a need for non-invasive diagnostic tools.

Purpose of the Study:

• To develop and evaluate a neural network model, Sequence Contrastive Learning (SeqCon), for differentiating PR from PE using only coronary angiography (CAG).
• To assess the feasibility of diagnosing ACS pathogenesis non-invasively through CAG analysis.

Main Methods:

• Utilized 842 videos from 278 ACS patients (172 PR, 106 PE), with ground truth confirmed by Optical Coherence Tomography (OCT).
• Developed SeqCon, a novel contrastive learning approach, to enhance feature learning from consecutive video frames for improved classification.
• Validated the model on an external test set comprising 18 PR and 11 PE patients.

Main Results:

• SeqCon achieved high patient-level performance: 82.8% accuracy, 88.9% sensitivity, 72.3% specificity, 84.2% positive predictive value, and 80.0% negative predictive value.
• Demonstrated the model's ability to distinguish between PR and PE effectively using only CAG data.
• This marks the first application of contrastive learning for diagnosing ACS mechanisms via CAG.

Conclusions:

• Coronary angiography, analyzed with advanced AI like SeqCon, can feasibly differentiate between plaque rupture and plaque erosion in ACS.
• This non-invasive approach holds potential for guiding clinical management decisions, potentially reducing the need for more invasive diagnostic procedures.