Deep Learning Predicts Cardiac Output from Seismocardiographic Signals in Heart Failure.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence

Background:

• Accurate cardiac output (CO) measurement is vital for managing cardiovascular compromise.
• Standard methods like right heart catheterization (RHC) are invasive and have limitations.
• Seismocardiography (SCG) offers a non-invasive alternative by sensing cardiac mechanical activity.

Purpose of the Study:

• To develop and evaluate a deep learning model for estimating CO using SCG, ECG, and BMI.
• To assess the model's performance in heart failure patients undergoing RHC.

Main Methods:

• A deep convolutional neural network was trained on an open-access dataset of 73 heart failure patients.
• Data included simultaneous RHC, SCG, and ECG recordings.
• Model performance was validated on 64 patients using cross-validation.

Main Results:

• The model achieved a mean bias of -0.01 L/min for CO (< 6 L/min) and 0.07 L/min/m² for cardiac index (< 2.2 L/min/m²).
• Limits of Agreement were -0.88 to 0.87 L/min for CO and -0.35 to 0.48 L/min/m² for cardiac index.
• Performance was notably strong in low-output states.

Conclusions:

• Deep learning with wearable SCG sensors can non-invasively estimate CO.
• SCG-based monitoring holds potential for clinical decision-making where invasive methods are impractical.
• Further multicenter validation is required to confirm generalizability.