Deep Learning for Cardiac Overload Estimation　- Predicting B-Type Natriuretic Peptide (BNP) Levels From Heart Sounds...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• B-type natriuretic peptide (BNP) and N-terminal pro-BNP (NT-pro-BNP) are crucial biomarkers for heart failure (HF) management.
• Traditional auscultation for HF lacks objective evaluation.
• Phonocardiogram devices can rapidly record heart sounds and ECG.

Purpose of the Study:

• To develop a deep-learning model for estimating plasma BNP levels from non-invasive heart sound and ECG signals.
• To validate the model's performance using an external dataset.
• To assess the clinical feasibility of this non-invasive approach for HF screening.

Main Methods:

• A deep-learning model was developed to estimate BNP levels using 8-second heart sound and ECG recordings.
• The model was evaluated on an external validation dataset of 140 patients.
• Subgroup analysis was conducted on patients with a specific body mass index range.

Main Results:

• The estimated BNP (eBNP) model achieved an area under the receiver operating characteristic curve (AUROC) of 0.895 for predicting BNP levels ≥100 pg/mL.
• Sensitivity and specificity were 84.3% and 82.9%, respectively.
• In a subgroup (BMI 18.5-25), the model showed enhanced predictive capability with an AUROC of 0.959, sensitivity of 92.5%, and specificity of 84.8%.

Conclusions:

• The eBNP model shows significant potential for non-invasive and rapid heart failure screening.
• Its objective and simple nature makes it suitable for point-of-care testing.
• The model's robustness across diverse populations, validated on independent datasets, supports its use for early HF diagnosis and monitoring.