Radiotherapy Dose Distribution Prediction Based on Machine Learning Using Random Forest Model in Breast Cancer Cases
Fany Zahra Hanifa, Dwi Seno Kuncoro Sihono, Amar Ma^ruf Irfan Muhamadi

Faculty of Mathematics and Natural Science, Universitas Indonesia
Department of Radiotherapy, MRCCC Siloam Hospital Semanggi

Abstract

Breast cancer is one of the cancers with the highest prevalence in the world and has become a significant global health threat. Radiotherapy using the Intensity-Modulated Radiation Therapy (IMRT) technique is a widely used treatment modality- however, the Treatment Planning System (TPS) process requires considerable time and expertise, making the development of Machine Learning (ML)-based models highly relevant. This study aims to implement the Random Forest algorithm to predict radiotherapy dose-volume parameters in breast cancer cases based on radiomics features extracted from patient CT imaging data. A total of 284 DICOM radiotherapy planning datasets were obtained from MRCCC Siloam Hospitals Semanggi, South Jakarta. Radiomics shape features served as input variables, while dosiomics parameters covering dose and volume of the target (PTV) and organs-at-risk (OAR) were used as output variables, normalized to each patient^s prescribed dose and PTV volume. The model was developed using Python with a 70% training and 30% testing split, optimized using RandomizedSearchCV. Evaluation was performed using MSE, RMSE, R-squared, MAE, and paired t-test. The model showed strong performance for OAR volume parameters (R-squared 0.82-0.89 for heart and lungs). For the primary target (PTV 50Gy), key dose parameters were predicted with an error of 0.27%-1.75% relative to the prescribed dose, within the typical 2%-5% clinical tolerance. No significant difference was found between predicted and clinical results for 10 of 11 key parameters (p > 0.05), including the Homogeneity Index (p = 0.170) and Conformity Index (p = 0.614). The Random Forest model is expected to serve as an effective tool in supporting faster and more standardized radiotherapy planning.

Keywords: Breast Cancer, Radiotherapy, Machine Learning, Random Forest, Dose Distribution

Topic: Medical Physics and Biophysics