Machine Learning-Based Prediction Models for Magnetic Hyperthermia

Magnetic hyperthermia using heat generation of magnetic nanoparticles (MNPs) has great potential as a noninvasive cancer treatment with minimal side effects. Despite its promising results from basic research, the limited success in clinical translation stems from a lack of systematic nanoparticle optimization. To bridge this gap, machine learning (ML) models that predict the heating efficiency of MNPs composed of different compositions from the physical and magnetic properties of nanoparticles and external magnetic field conditions were developed in this study. The comparative analysis showed that the artificial neural network (ANN), gradient boosting regressor, and random forest regressor outperformed compared to support vector machine (SVM) and k-nearest neighbor (KNN). The ANN models with two hidden layers showed good prediction performances comparative to deeper ANN models. Furthermore, the feature importance analysis revealed the contributions of each input features toward the heating efficiency of MNPs based on the real experimental data.