Prediction of Radiotherapy Dose Distribution for Glioblastoma Using Convolutional Neural Network Model
Ni Made Dian Rani
Department of Physics, Faculty of Mathematics and Natural Sciences, Udayana University, Jimbaran Campus, Badung, 80361, Bali, Indonesia.
Ni Nyoman Ratini
Department of Physics, Faculty of Mathematics and Natural Sciences, Udayana University, Jimbaran Campus, Badung, 80361, Bali, Indonesia.
Anak Agung Ngurah Gunawan *
Department of Physics, Faculty of Mathematics and Natural Sciences, Udayana University, Jimbaran Campus, Badung, 80361, Bali, Indonesia.
Gusti Ngurah Sutapa
Department of Physics, Faculty of Mathematics and Natural Sciences, Udayana University, Jimbaran Campus, Badung, 80361, Bali, Indonesia.
Hery Suyanto
Department of Physics, Faculty of Mathematics and Natural Sciences, Udayana University, Jimbaran Campus, Badung, 80361, Bali, Indonesia.
I Gde Antha Kasmawan
Department of Physics, Faculty of Mathematics and Natural Sciences, Udayana University, Jimbaran Campus, Badung, 80361, Bali, Indonesia.
*Author to whom correspondence should be addressed.
Abstract
Aims: This study aims to predict radiotherapy dose distribution for glioblastoma patients using Machine Learning with a Convolutional Neural Network (CNN) model.
Study Design: This research used an experimental design with a quantitative approach to predict radiotherapy dose distribution for glioblastoma patients using a CNN model. The study involved training and testing the CNN model on medical imaging data from The Cancer Imaging Archive (TCIA), evaluating its performance based on Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Structural Similarity Index Measure (SSIM), Dice Similarity Coefficient (DSC), Peak Signal to Noise Ratio (PSNR), Normalized Cross-Correlation (NCC). The results were analyzed to determine the model’s accuracy in replicating actual dose distributions, providing a data-driven assessment of its predictive capability.
Place and Duration of Study: This research was conducted in the Department of Physics at Udayana University from October 2024 to January 2025.
Methodology: The research involved 180 patient datasets divided into 126 training data and 54 testing data. The CNN architecture is implemented using the Google Collaboratory platform. Model evaluation is performed using MSE, RMSE, and SSIM to measure the accuracy of dose distribution prediction.
Results: The MSE, RMSE, SSIM, DSC, PSNR, and NCC values obtained from the CNN model are 0.00015795, 0.01256, 0.979718, 0.9711, 32dB, and 0.96289 respectively. The low MSE and RMSE values indicate minimal prediction error, while the high SSIM confirms strong structural similarity between the predicted and actual dose maps. The DSC demonstrates excellent spatial overlap, and the high PSNR reflects high-quality dose reconstruction. Additionally, the NCC highlights strong correlation with the ground truth. Visually, the axial, coronal, and sagittal slices closely resemble the actual dose distributions, further validating the model’s accuracy.
Conclusion: The CNN model demonstrates effectiveness in predicting the dose distribution for glioblastoma radiotherapy, achieving highly accurate evaluation metrics. Visually, the model exhibit patterns highly similar to the actual dose map.
Keywords: CNN, dose distribution, glioblastoma, radiotherapy, machine learning