Artificial intelligence aided microwave coagulation therapy: Analysis of heat transfer to tumor tissue via hybrid modeling

Abstract

This study investigates the application of three regression models including Decision Tree (DT), Light Gradient Boosting Machine (LGBM), and Random Forest (RF) to predict temperature distribution in a tumor tissue based on two inputs, r(m) and z(m), using a dataset comprising over 64,000 records. Temperature distribution near the tumor tissue was calculated using numerical simulation of heat transfer which constituted electromagnetic and bioheat transfer models. The simulations were carried out for cancer treatment via hyperthermia utilizing antenna for electromagnetic heating. Machine learning models were trained in combination with computational heat transfer to obtain temperature distribution. We employed Massively Parallel Hyperparameter Tuning (MPPT) to optimize the hyperparameters for each model, ensuring optimal performance. Obtaining a score of 0.9991 by R² criterion (Coefficient of Determination), an MSE (Mean Squared Error) of 0.1526, and an MAE (Mean Absolute Error) of 0.2545, the results show that the LGBM is the best-fit model. With R² = 0.9981, MSE = 0.3039, and MAE = 0.2744, the Random Forest model was likewise quite effective. The Decision Tree model still performed adequately, but it was not as effective as the other options (R² = 0.9923, MSE = 1.2868, MAE = 0.6138). Hyperparameter tuning is crucial for obtaining better model performance, and these results show that LGBM works well for regression tasks on big datasets. The results indicated that machine learning is a robust technique in evaluation of cancer therapy efficiency via hyperthermia method by determination of temperature distribution in the tumor tissue.