DoseGNN:通过图神经网络提升自适应剂量-体积直方图预测中深度学习模型的性能
DoseGNN: Improving the Performance of Deep Learning Models in Adaptive Dose-Volume Histogram Prediction through Graph Neural Networks
AI总结:
针对自适应放射治疗中DVH预测的图像异质性问题,提出基于GNN的即插即用DoseGNN框架,可提升各类基础深度学习模型在配备CBCT的通用放疗平台上的DVH预测性能。
摘要:
剂量-体积直方图(Dose-Volume Histogram, DVH)预测是放射治疗中的基础环节,可辅助治疗计划制定、剂量评估、方案比较等工作。它有助于提升精准有效放射治疗的实施能力,同时根据需要管控对健康组织的潜在毒性,以降低并发症风险。本文拓展了此前在AAPM(美国医学物理学家协会第65届年会暨展览会)上公布的研究成果,并补充了必要的技术细节。研究目标是为配备高性能CBCT系统的通用放射治疗平台设计高效的DVH预测深度学习模型,该场景下输入的CT图像与待预测的靶区剂量图像可能存在来源、间距及尺寸差异。研究在该新型放射治疗平台上评估了DVH预测任务中广泛采用的深度学习模型,结果表明图神经网络(GNNs)是构建即插即用框架的理想架构,可提升基础深度学习模型在自适应场景下的预测性能。
英文摘要:
Dose-Volume Histogram (DVH) prediction is fundamental in radiation therapy that facilitate treatment planning, dose evaluation, plan comparison and etc. It helps to increase the ability to deliver precise and effective radiation treatments while managing potential toxicities to healthy tissues as needed to reduce the risk of complications. This paper extends recently disclosed research findings presented on AAPM (AAPM 65th Annual Meeting $\&$ Exhibition) and includes necessary technique details. The objective is to design efficient deep learning models for DVH prediction on general radiotherapy platform equipped with high performance CBCT system, where input CT images and target dose images to predict may have different origins, spacing and sizes. Deep learning models widely-adopted in DVH prediction task are evaluated on the novel radiotherapy platform, and graph neural networks (GNNs) are shown to be the ideal architecture to construct a plug-and-play framework to improve predictive performance of base deep learning models in the adaptive setting.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。