HCM-Echo-VAR-Ensemble: Deep Ensemble Fusion to Detect Hypertrophic Cardiomyopathy in Echocardiograms

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Hypertrophic Cardiomyopathy (HCM) is a significant cardiovascular disease requiring accurate diagnostic methods.
• Echocardiogram videos are crucial for assessing cardiac structure and function.
• Current diagnostic approaches may have limitations in sensitivity and consistency.

Purpose of the Study:

• To develop and validate a novel deep learning framework for automated Hypertrophic Cardiomyopathy detection.
• To leverage advanced video analysis models for improved diagnostic accuracy from echocardiograms.
• To establish a robust and reliable method for HCM classification using cardiac ultrasound.

Main Methods:

• Proposed HCM-Echo-VAR-Ensemble, a framework utilizing an ensemble of deep Video-Action-Recognition (VAR) architectures (SlowFast and I3D).
• Employed majority averaging ensembling to fuse predictions from individual VAR models for binary classification (HCM vs. no HCM).
• Directly analyzed echocardiogram videos for HCM detection without manual feature extraction.

Main Results:

• Achieved state-of-the-art accuracy of 95.28% and an AUC of 98.42%.
• Demonstrated high performance metrics including F1-Score (95.20%), specificity (96.20%), sensitivity (93.97%), PPV (96.46%), and NPV (94.17%).
• Outperformed various baseline methods and other ensembling approaches.

Conclusions:

• The HCM-Echo-VAR-Ensemble framework shows significant potential for enhancing HCM detection accuracy and sensitivity in clinical practice.
• Ensembling complementary strengths of SlowFast and I3D models improves diagnostic consistency.
• This approach offers reliable HCM diagnosis, particularly beneficial in low-resource environments.