CasMamDiff: Cascaded Mamba Diffusion Model for Radiotherapy Dose Prediction

Dose prediction is essential for the planning of radiotherapy treatment in cancer care. In particular, accurate and efficient dose prediction can enhance the planning workflow and improve treatment quality. Today, with the application of deep learning to radiation therapy planning, the quality and efficiency of dose prediction are improving. However, existing approaches based on convolutional neural networks (CNNs) or Transformers face notable limitations. CNNs lack the capacity to capture global dependencies, while Transformers are computationally expensive. To address these problems, we introduce the Cascaded Mamba Diffusion Model, a model that combines Mamba and a cascaded architecture. Furthermore, we incorporate a 3D distance map to provide anatomical spatial guidance. To validate the performance, our model is evaluated on the OpenKBP dataset with a total of 340 patient cases. Through our experiment, our model demonstrates superior performance over other state-of-the-art methods. This demonstrates our method's potential in improving radiotherapy dose prediction accuracy and ensuring better treatment planning efficiency.