An Innovative Artificial Intelligence Classification Model for Non-Ischemic Cardiomyopathy Utilizing Cardiac...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Non-ischemic cardiomyopathies (NICMs) present diagnostic challenges due to overlapping morphology and subtle functional changes.
• Current AI models for cardiac magnetic resonance (CMR) often miss crucial biomechanical dysfunctions.
• Intraventricular pressure gradients (IVPGs) offer valuable biomechanical insights often overlooked.

Purpose of the Study:

• To develop and validate a novel dual-path hybrid deep learning framework for improved NICM diagnosis and subtype classification.
• To integrate anatomical CMR data with biomechanical IVPG markers for enhanced diagnostic accuracy.
• To capture subtle biomechanical dysfunctions missed by morphology-based AI models.

Main Methods:

• A dual-path hybrid deep learning model combining CNN-LSTM for cine CMR and MLP for IVPG time-series data was developed.
• The framework was trained on a multicenter dataset (1196 patients) and externally validated (137 patients).
• Model performance was evaluated using AUC, compared against ResNet50, VGG16, and radiomics-SVM, with ablation studies and visualization techniques (Grad-CAM).

Main Results:

• The proposed model achieved superior performance with high AUC values (internal: 0.974, external: 0.962).
• IVPGs were confirmed as a significant contributor to diagnostic accuracy.
• The model demonstrated robust generalizability across different imaging protocols and institutions.

Conclusions:

• The dual-path deep learning framework effectively integrates biomechanical IVPG data with CMR imaging for superior NICM classification.
• This approach offers an interpretable and data-efficient solution for early NICM detection and subtype differentiation.
• The model shows strong potential for clinical translation in diagnosing complex cardiomyopathies.