Multimodal deep learning enhances diagnostic precision in left ventricular hypertrophy

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Differentiating causes of left ventricular hypertrophy (LVH) is clinically challenging due to similar presentations.
• Accurate diagnosis of hypertrophic cardiomyopathy (HCM) versus hypertension (HTN) is crucial for patient management and family screening.
• Current diagnostic methods may require augmentation for improved accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate an AI method, LVH-fusion, for joint interpretation of ECGs and echocardiogram videos.
• To augment physician interpretation in distinguishing HCM from HTN.
• To improve diagnostic accuracy for LVH aetiologies, particularly HCM.

Main Methods:

• Developed LVH-fusion using over 18,000 ECGs and echocardiograms from 2,728 patients.
• Trained the AI on electronic health record blood pressure measurements and expert diagnostic consensus, not physician over-reads.
• Utilized explainability techniques to identify key diagnostic features for HCM and HTN.

Main Results:

• LVH-fusion achieved an F1-score of 0.71 for predicting HCM and 0.96 for predicting HTN on held-out test data.
• The AI demonstrated higher sensitivity and specificity compared to human readers in head-to-head comparisons.
• Explainability analysis confirmed the diagnostic importance of lateral T-wave inversion (ECG) and proximal septal hypertrophy (echocardiogram) for HCM.

Conclusions:

• Deep learning models, like LVH-fusion, can effectively augment physician capabilities in diagnosing complex cardiac conditions.
• This AI-driven approach has significant implications for preventing sudden cardiac death by enabling earlier and more accurate diagnoses.
• LVH-fusion offers a promising tool for addressing diagnostic challenges in hypertrophic cardiomyopathy and hypertension.