Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Unsupervised deep learning model for fast energy layer pre-selection of delivery-efficient proton arc therapy plan optimization of nasopharyngeal carcinoma

Created by
  • Haebom

Author

Bohan Yang, Gang Liu, Yang Zhong, Rirao Dao, Yujia Qian, Ke Shi, Anke Tang, Yong Luo, Qi Kong, Jingnan Liu

Outline

This paper proposes SPArc_dl, an unsupervised deep learning model for rapid preselection of optimal energy layer (EL) sequences in proton beam therapy (PAT). To address the computational burden and extended treatment times compared to existing methods, we introduce a novel data representation, the spot count representation, which encodes the number of proton spots passing through the target and organs-at-risk (OAR). SPArc_dl, a U-Net architecture, is trained with a triple objective function: maximizing the number of target spots, minimizing the number of OAR spots, and reducing the EL switching time. Evaluation results using data from 35 pharyngeal cancer patients show that SPArc_dl improves both planning quality and delivery efficiency compared to the existing method, SPArc_ps. Specifically, it improves the fitness index by 0.1, reduces the homogeneity index by 0.71, reduces the mean brainstem dose by 0.25, and shortens the EL switching time by 37.2%. The inference time is less than 1 second, and we show that not changing the order of EL transitions is more time-efficient. However, the SPArc_dl plan shows limitations in robustness.

Takeaways, Limitations

•
Takeaways:
◦
We demonstrate that unsupervised deep learning can effectively solve the energy layer preselection problem in PAT.
◦
We propose a new data representation method called spot count representation to improve the performance of deep learning models.
◦
SPArc_dl is a fast tool that improves planning quality and reduces treatment time compared to existing methods.
◦
We found that not changing the EL transition order was more time efficient.
•
Limitations:
◦
Lack of robustness of SPArc_dl plan.
◦
Because we evaluated the model using only data from a limited type of cancer (pharyngeal cancer), its generalization performance to other cancer types requires further study.
👍
Made with Slashpage