Dose prediction of intensity-modulated radio therapy for brain tumors based on an improved deep learning framework

Accurate calculation of dose distribution is a critical component in intensity-modulated radiation therapy (IMRT). The dose distribution exhibits significant variability across patients. To address this challenge, we propose a deep learning-based dose prediction method specifically designed for brain tumors in the context of IMRT. By incorporating the anatomical characteristics of brain tumors, we developed an innovative convolutional neural network architecture. Furthermore, we employed a structural similarity loss function to enhance the precision of dose distribution predictions for brain tumor patients. Experimental results demonstrate its superior performance over state-of-the-art methods in terms of Mean Absolute Error (MAE) and Dose Volume Histogram (DVH) metrics.