Prediction of Left Atrial Volume Parameters from Resting ECGs and Tabular Data Using Deep Learning in the UK Biobank.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Left atrial (LA) volume is a critical indicator of cardiovascular health.
• Magnetic Resonance Imaging (MRI) is the gold standard for LA volume measurement but is costly and not widely accessible.
• There is a need for scalable, low-cost methods for LA volume assessment.

Purpose of the Study:

• To develop and validate a deep learning model for predicting left atrial (LA) volume.
• To utilize standard 12-lead electrocardiogram (ECG) recordings and basic patient data as input.
• To provide an interpretable and cost-effective alternative to MRI-based LA volume quantification.

Main Methods:

• A deep learning model was trained to perform regression on LA volume targets.
• The model incorporated standard 12-lead ECG data and basic patient demographics (e.g., weight, height).
• Shapley values were employed to interpret feature importance and understand model predictions.

Main Results:

• The deep learning model accurately predicted LA volume from ECG and patient data.
• ECG signals were identified as significant predictors of LA volume.
• Patient features like weight and height were found to contribute meaningfully to the volume estimation.

Conclusions:

• Deep learning models can effectively predict left atrial (LA) volume using readily available ECG and patient data.
• This approach presents a scalable and low-cost alternative to MRI for LA volume assessment.
• The model's interpretability through Shapley values enhances clinical trust and understanding.