A Novel Approach to Identifying Hibernating Myocardium Using Radiomics-Based Machine Learning

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Assessing myocardial viability is crucial for managing ischemic heart disease.
• Hibernating myocardium, characterized by impaired contractility despite reduced perfusion, requires accurate detection.
• Current methods may have limitations in fully characterizing myocardial viability.

Purpose of the Study:

• To evaluate the feasibility of using machine learning (ML) with radiomics features from rest myocardial perfusion imaging (MPI) to detect hibernating myocardium.
• To explore the potential of ML in identifying subtle perfusion defects indicative of hibernating myocardium.

Main Methods:

• Retrieved data from patients who underwent 99mTc-sestamibi MPI and 18F-FDG PET/CT.
• Extracted 110 radiomics features from perfusion defects on MPI polar maps.
• Trained and validated 13 supervised ML algorithms, selecting the best performing models based on Log Loss and Area Under the Curve (AUC).
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Nine ML models achieved an AUC > 0.800 on the unseen testing set.
• Gradient Boosting Random Forest (xgboost) demonstrated the highest AUC of 0.860.
• The best model detected hibernating myocardium in 72.4% of defects with 78.0% classification accuracy and 0.792 F1 Score.
• Four models showed clear interpretability via SHAP beeswarm plots.

Conclusions:

• Machine learning applied to radiomics features from rest MPI is a feasible approach for detecting hibernating myocardium.
• Radiomics features capture information not apparent to the human eye, enhancing myocardial viability assessment.
• This proof-of-concept study highlights a promising avenue for improving cardiac patient management.