Fluence Map Prediction with Deep Learning: A Transformer-based Approach

The recent study on fluence map prediction in intensity-modulated radiation therapy (IMRT) highlights a transformative approach using a deep learning framework based on a 3D Swin-UNETR network. By training on 99 prostate IMRT cases, the model demonstrated impressive performance metrics, including an average R^2 of 0.95 and a gamma passing rate of 85%. This method allows for fully automated and inverse-free fluence map predictions directly from anatomical inputs, significantly enhancing the efficiency and accuracy of treatment planning. The model's predictions were evaluated against clinical plans, showing no significant differences in dose-volume histogram (DVH) parameters, thereby confirming its clinical viability. This innovation not only accelerates the planning process but also ensures that tumor coverage is maximized while minimizing exposure to healthy tissues, which is crucial for improving patient outcomes in radiation therapy.