Feature Mapping of Native Oxygenation-Sensitive CMR Images for Classifying Cardiomyopathies.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cardiovascular disease is a major global health issue.
• Innovative, needle-free diagnostic tools are crucial for early detection and management.
• Oxygenation-sensitive cardiovascular magnetic resonance (OS-CMR) imaging offers potential for non-invasive assessment.

Purpose of the Study:

• To integrate OS-CMR imaging with deep learning for myocardial pathology classification.
• To evaluate the diagnostic performance of the combined approach across ischemic, non-ischemic, edema, and healthy myocardium.
• To assess the interpretability of the deep learning model using AI-derived feature maps.

Main Methods:

• A deep learning model was trained on OS-CMR images from 180 patients (4 categories).
• Stratified 5-fold cross-validation with Monte Carlo Dropout and residual learning was employed.
• AI-derived feature maps were generated and compared with expert segmentations (LGE, T2 maps).

Main Results:

• The model achieved high AUC scores: 0.93 (healthy), 0.80 (ischemic), 0.89 (non-ischemic), and 0.96 (edema).
• AI-derived maps showed strong spatial correlation with expert-defined lesions (Dice values up to 0.93).
• OS-CMR data contains latent information for deep learning-based diagnostic classification.

Conclusions:

• Deep learning applied to OS-CMR imaging effectively classifies myocardial pathologies.
• The approach offers a promising needle-free, ultra-efficient diagnostic tool for cardiovascular disease.
• AI interpretability methods confirm the model's ability to identify relevant pathological features.