使用生成对抗网络预测头癌放射治疗的剂量分布:输入数据的影响
Xiaojin Gu1,2, Victor I J Strijbis1,2, Ben J Slotman1,2
1Department of Radiation Oncology, Amsterdam UMC Location Vrije Universiteit Amsterdam, Amsterdam, Netherlands.
Frontiers in oncology
|October 13, 2023
概括
深度学习模型可以在没有完全细分的情况下预测头癌的放射治疗剂量分布. 包括CT扫描和与有风险的器官 (OARs) 规划目标体积 (PTVs) 在内,提供了准确的预测,减少了手动细分需求.
科学领域:
- 医学物理 医学物理
- 辐射疗法 辐射疗法
- 人工智能的人工智能
背景情况:
- 计划放射治疗的目标体积 (PTV) 和风险器官 (OAR) 的手动细分是耗时的.
- 准确的剂量分配预测对于有效的癌症治疗计划至关重要.
研究的目的:
- 调查使用3D深度生成对抗网络 (GAN) 预测剂量分布的可行性,而不需要完全细分的数据集.
- 评估不同输入数据组合 (CT扫描,PTV,OAR) 对剂量预测准确性的影响.
主要方法:
- 在320个CT数据集和治疗计划上训练并验证了3D GAN.
- 使用各种输入组合测试的模型:CT-only (C),CT+PTVs (CP),CT+PTVs+OARs+body (CPOB),PTVs+OARs+body (POB),PTVs+body (PB) 等.
- 对剂量分配和OARs的平均剂量分析的平均绝对误差 (MAE).
主要成果:
- 包括PTV轮在内的模型与临床计划有很高的一致性. 在CP,CPOB,POB和PB模型中,MAE范围为3.4-3.5 Gy,而仅CT (C) 则为7.3 Gy.
- CT+PTVs (CP) 模型的准确度与CT+PTVs+OARs+body (CPOB) 模型的准确度相似,表示最小的数据要求.
- 对OAR平均剂量的预测与包括PTV轮 (R2值0.95-0.97) 在内的模型的临床分布高度相关.
结论:
- CT扫描和PTV轮显著影响剂量预测的准确性.
- CT+PTVs (CP) 模型代表了在患者队列中充分预测临床剂量分布所需的最低数据.
- 深度学习模型可以通过减少对广泛的手动细分的依赖来简化放射治疗计划.
相关概念视频
Dose-Response Relationship: Overview
3.2K
Agonists can bind with and activate receptors, resulting in the formation of drug-receptor complexes. Once formed, these complexes catalyze many biochemical processes at the cellular level and subsequently induce a pharmacologic response. The degree of response is directly proportional to the fraction of activated receptors, which in turn, depends on the concentration of the drug at the receptor site as well as the sensitivity of the receptor. An increase in the administered dose contributes to...
3.2K
Dose-Response Relationship: Potency and Efficacy
4.5K
The potency of a drug is the measure of its ability to produce a biological response and can be compared by looking at the half-maximum effective concentration or EC50 values of different drugs. A lower EC50 value indicates higher potency of the drug. In the dose–response curve of two antihypertensive drugs, candesartan and irbesartan, a significant difference is observed in their EC50 values. A lower EC50 value for candesartan indicates that it is more potent than irbesartan, as it...
4.5K


