Predicting left ventricular remodeling post-MI through coronary physiological measurements based on computational...

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Early detection of left ventricular remodeling (LVR) is critical for managing patients post-myocardial infarction.
• Cardiac magnetic resonance (CMR) is a standard for LVR assessment but has limitations.
• Coronary angiography-derived parameters like fractional flow reserve (caFFR) and index of microcirculatory resistance (caIMR) offer potential alternatives.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting LVR in ST-segment elevation myocardial infarction (STEMI) patients.
• To compare the predictive performance of models incorporating clinical, CMR, and coronary physiology parameters.
• To identify key predictors of LVR using SHAP analysis.

Main Methods:

• Prospective inclusion of 157 STEMI patients undergoing primary percutaneous coronary intervention.
• Development of three random forest machine learning models: Model 1 (clinical/procedural), Model 2 (Model 1 + CMR), and Model 3 (Model 1 + echocardiographic/functional parameters including caFFR and caIMR).
• Assessment of model performance using Area Under the Curve (AUC) and feature importance via SHAP analysis.

Main Results:

• Left ventricular remodeling (LVR) was present in 23.6% of patients.
• Model 3, incorporating coronary physiology (caFFR, caIMR) and echocardiographic data, achieved the highest AUC (0.85), followed by Model 2 (0.84) and Model 1 (0.77).
• Key predictors identified included infarct size, microvascular obstruction, admission hemoglobin, current smoking, and caFFR.

Conclusions:

• Coronary physiology parameters (caFFR, caIMR) combined with echocardiography effectively predict LVR in STEMI patients.
• These parameters demonstrate strong predictive performance, comparable or superior to CMR-based models.
• The findings suggest that caFFR and caIMR may serve as valuable, potentially non-invasive, alternatives to CMR for early LVR detection.