Detecting heart failure using novel bio-signals and a knowledge enhanced neural network

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Biomedical Signal Processing

Background:

• Clinical decisions for heart failure (HF) often rely on left ventricular ejection fraction (LVEF) measured by echocardiography.
• Echocardiography is unavailable in primary care, delaying HF detection and leading to prolonged diagnostic processes.
• Standard 12-Lead ECG has limited sensitivity and specificity for detecting structural and functional heart abnormalities.

Purpose of the Study:

• To evaluate the effectiveness of an AI solution, Cardio-HART™ (CHART), in reducing the diagnostic gap for heart disease in primary care.
• To assess CHART's performance in predicting heart failure (HF) compared to existing ECG-based criteria.
• To enable early detection of structural, functional, and valve abnormalities in primary care settings.

Main Methods:

• Utilized Knowledge-enhanced Neural Networks to process novel bio-signals alongside ECG data.
• Applied the Cardio-HART™ (CHART) AI algorithm to predict heart abnormalities.
• Compared CHART's diagnostic performance against established ECG-based criteria for heart failure.

Main Results:

• CHART demonstrated significantly improved performance in HF prediction: sensitivity increased from 53.5% to 82.8%, specificity from 85.1% to 86.9%.
• The AI solution doubled the sensitivity for HF-indicated findings (38.6% to 71%) while maintaining similar specificity.
• CHART achieved a higher F-score (72.2% vs. 56.4%) and area under the curve (0.91 vs. 0.79) compared to ECG criteria.

Conclusions:

• The AI-powered CHART algorithms can predict structural, functional, and valve abnormalities using ECG and novel bio-signals.
• CHART effectively reduces the diagnostic gap, facilitating early detection of heart diseases and HF in primary care.
• This AI solution offers a promising approach for timely diagnosis and intervention in primary care settings.