ResFouriONet: A Residual Fourier Operator Network with Synthetic Data Generation for Real-Time Laser-Induced Bioheat...

Area of Science:

• Biomedical Engineering
• Computational Physics
• Medical Imaging

Background:

• Accurate modeling of laser-tissue interactions is crucial for medical applications like brain imaging and laser therapies.
• Traditional methods (e.g., Monte Carlo simulations) are computationally intensive, hindering real-time applications.
• Operator learning offers scalability but faces challenges with extrapolation and data generation.

Purpose of the Study:

• To develop a computationally efficient deep operator network for modeling 2D transient bioheat transfer with parametric laser sources.
• To introduce a novel synthetic data generation scheme to reduce dependence on costly Monte Carlo simulations.
• To enable accurate, real-time prediction of laser-induced thermal effects.

Main Methods:

• Introduced ResFouriONet, a residual deep operator network with a dual-pathway architecture (Fourier and cascaded residual sub-branches) and self-attention.
• Developed a synthetic laser source function generation scheme using Gaussian profiles, exponential decay, and random perturbations.
• Evaluated performance on unseen synthetic data and Monte Carlo-generated source functions.

Main Results:

• ResFouriONet achieved a 5x reduction in prediction error compared to vanilla deep operator networks.
• Attained an average relative error of 0.93% on synthetic data and 1.45% on Monte Carlo-generated data.
• Demonstrated accurate, real-time prediction capabilities with fewer parameters.

Conclusions:

• The proposed ResFouriONet architecture and data generation strategy enable efficient and accurate modeling of laser-induced heating.
• This facilitates simulation-free planning and control in biomedical imaging and laser therapies.
• Paves the way for real-time thermal management in laser-based medical procedures.