Physics-driven Spatiotemporal Regularization for High-dimensional Predictive Modeling: A Novel Approach to Solve the...

Area of Science:

• Biomedical Engineering
• Computational Science
• Medical Imaging

Background:

• Predictive modeling in complex healthcare systems requires capturing intricate spatiotemporal dynamics.
• Existing regularization methods often struggle with high-dimensional data and physics-based interrelationships.

Purpose of the Study:

• To introduce a novel physics-driven spatiotemporal regularization (STRE) method for improved high-dimensional predictive modeling.
• To enhance the prediction of time-varying spatial distributions of physiological variables.

Main Methods:

• Developed a STRE model integrating physics-based relationships between explanatory and response variables.
• Applied spatial and temporal regularization to improve prediction accuracy.
• Utilized electrocardiogram (ECG) data from body surface sensors to predict heart surface electric potentials.

Main Results:

• The STRE model demonstrated superior performance in predicting electric potentials on the heart surface.
• Validation in both simulated and realistic torso-heart geometries confirmed the model's efficacy.
• STRE significantly outperformed Tikhonov zero-order, Tikhonov first-order, and L1 first-order regularization methods.

Conclusions:

• The physics-driven STRE method offers a significant advancement in predictive modeling for healthcare.
• STRE provides a robust framework for analyzing complex spatiotemporal data, particularly in electrocardiology.
• This approach holds promise for improving diagnostic and prognostic capabilities in medical applications.