Lightweight preprocessing and template matching facilitate streamlined ischemic myocardial scar classification.

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Ischemic myocardial scarring (IMS) is a critical complication of coronary artery disease, leading to arrhythmias and heart failure.
• Late-gadolinium-enhanced cardiac magnetic resonance (CMR) imaging is the standard for IMS diagnosis.
• Machine learning (ML) improves IMS classification accuracy but requires substantial computational resources.

Purpose of the Study:

• To introduce and evaluate a streamlined approach for IMS classification using lightweight preprocessing (LWP) and template matching (TM).
• To assess the efficiency and diagnostic performance of LWP combined with TM compared to deep learning methods (CNNs).

Main Methods:

• Cardiac MRI scans from 279 patients were analyzed using two CNNs and TM, with and without LWP.
• Performance was evaluated using accuracy, sensitivity, specificity, F1-score, AUROC, and processing time.
• External datasets were used for patient-level classification and direct comparison between CNNs and TM.

Main Results:

• LWP accelerated CNNs by 4.9x and TM by 21.9x.
• TM was over 100x faster than CNNs when both utilized LWP.
• TM demonstrated comparable accuracy, sensitivity, specificity, F1-score, and AUROC to CNNs.
• Patient-level classifications showed significant metric improvements with the proposed methods.

Conclusions:

• LWP and TM effectively streamline IMS classification, offering a computationally efficient alternative to CNNs.
• Future enhancements in LWP and TM are expected to further improve diagnostic accuracy.
• This approach supports the potential for faster and more accessible IMS diagnosis.