PINNs Bioheat Experiments Repository

Overview

This repository contains code, models, and results for the observation of bioheat transfer problems using Physics-Informed Neural Networks (PINNs) and comparing their performance with ground truth simulations. The project is organized into three main folders: models, tests, and src.

Folder Structure

models/

This folder contains trained PINNs models, each uniquely identified by a random hash. For each model, the following files are provided:

• A trained model file.
• A plot showing the evolution of the loss components during training.
• A corresponding .yaml configuration file that defines the experiment setup. The unique hash is generated from the values in config.model_properties, so these values are the ones that characterize each model.

tests/

This folder contains the results of the experiments, organized into three categories:

1. cooling_simulation/:
   Contains both MATLAB ground truth and PINNs predictions:

   • 1obs: PINNs validation with a single observer (comparison with MATLAB ground truth).
   • 8obs: PINNs with a multiple-model observer, designed to outperform MATLAB.

2. cooling_meas_1/ and cooling_meas_2/:
   These folders contain the PINNs predictions on measurements performed at AMC Hospital.

3. inverse_problem_cooling_meas_1/ and inverse_problem_cooling_meas_2/:
   Results of inverse problem ran on measurements dataset to retrieve the perfusion value.

src/

Contains the source code for running experiments. Main scripts and folders:

• run.py: The main script to execute the complete set of experiments.
• main.py: Script for running individual experiments. Before running main.py, modify the experiment parameters in the config.yaml file located in the src folder.
• configs/: Contains predefined .yaml configuration files for specific experiments. Suffix "enhanced" stands for simulations with increased lambda and upsilon ( the two parameters of the ODE whose too much high value cause MATLAB to stop).
• matlab/: Contains MATLAB scripts used to generate ground truth simulations. Other utility files:
• coeff_calc.py: The script to load properties from config.yaml and calculate the coefficients of the PBHE.
• common.py: The script to perform some common functions, such as create output folders.
• ground_truth.py: The script to run matlab ground truth.
• import_data.py: The script to import measurements.
• measurements.py: The script to run prediction on measurements.
• plots.py: The script for plots.
• simulation.py: The script to run prediction on simulations.
• data/: Contains measurement data used in the experiments.
• heating/: Contains scripts about the antenna characterization.
• optimization/: Contains scripts for optimizing neural networks.
• pre-processing/: Contains scripts for measurements import.

Additional Files

Dockerfile is currently not updated.

requirements.txt

This file lists all the Python dependencies required to run the experiments. You can install them by running the following command:

pip install -r requirements.txt