Основні поняття
Automated dose prediction using MD-Dose based on the Mamba architecture enhances radiation therapy planning efficiency and accuracy.
Анотація
1. Introduction
Radiation therapy crucial in cancer treatment, requires precise plans.
Challenges include anatomical changes, collaboration delays, and plan optimization.
Automated dose prediction essential for expediting treatment process.
2. Diffusion Models in Dose Prediction
Deep learning used for automated dose distribution map prediction.
Diffusion models show potential in dose prediction without prior data knowledge.
Transformer-based UNet outperforms CNN-based models.
3. Mamba Architecture in MD-Dose
MD-Dose introduces a novel diffusion model based on the Mamba architecture.
Forward process adds noise to dose maps, backward process predicts noise to output accurate maps.
Extensive experiments showcase MD-Dose's superiority in metrics and time consumption.
4. Methodology
MD-Dose framework includes forward and reverse processes for noise addition and denoising.
Score-based diffusion generative models form the basis of MD-Dose design.
Mamba-based denoising network utilizes SSM layer and Mamba blocks for feature extraction.
5. Experiments & Results
Evaluation on clinical dataset of 300 thoracic tumor patients shows MD-Dose outperforms existing methods in various metrics.
Visual comparisons demonstrate MD-Dose's superior performance with minimal dose errors compared to ground truth.
6. Conclusion
MD-Dose offers efficient radiation dose prediction using the Mamba architecture, aiding in precise treatment planning for cancer patients.
Статистика
MD-Doseは、300人の胸部腫瘍患者のデータセットで優れた結果を示しました。
MD-Doseは、Dose Scoreが1.980、DVH Scoreが1.572、HI指数が0.285でした。