鼻咽癌是我国常见的恶性肿瘤，光子调强放疗是其标准放疗手段。临床上常采用自适应放疗技术去补偿器官结构改变导致的剂量变化。腮腺吸收过多剂量会导致体积缩小甚至引起严重并发症。但是吸收剂量同腮腺形状变化之间的关系尚不明确，临床上也缺乏精准保护腮腺的策略。申请人前期采用主成分分析方法实现了用较少成分准确重建出三维物理剂量，该方法同样适用于重建治疗后腮腺的形状变化，亦发现减小腮腺剂量能降低口干症的发生率。因此推测吸收剂量与腮腺几何变化之间存在关系，制定针对腮腺保护的放疗计划能维持腮腺体积并降低口干症的发生率。本项目拟利用200例(150训练，50临床推广)鼻咽癌患者自适应放疗的数据，通过机器学习的方法，1.建立吸收剂量与腮腺几何变化的关系用于预测; 2.以预测的腮腺几何变化为反馈，制定精准保护腮腺的剂量约束策略。本项目的实施有助于在鼻咽癌放疗中精准减小腮腺吸收剂量，维持腮腺体积，降低口干症的发生概率。

Nasopharyngeal cancer (NPC) is prevalent among populations in China and the standard treatment strategy is intensity modulated radiotherapy(IMRT). The adaptation was usually implemented to compensate for the dose variation caused by the geometric changes of organ in the clinic. Excessive doses absorbed by the parotid gland can lead to volume reduction and even serious complications.However, the relationship between dose distribution and the shape changes of parotid was undefined. There was a lack of the strategy for parotid sparing accurately as well. The three-dimensional dose was reconstructed using principal component analysis by the applicant previously. The shape changes of parotid during treatment could be reconstructed by the same method. In addition, it was found that the probability could be decreased by reducing the mean dose of the parotid. Therefore, it is supposed that there was a relationship between dose absorbed by parotid and the shape change of parotid. It is also supposed that the treatment plan with parotid sparing could reduce the probability of xerostomia. Based on the adaptive treatment data of 200 NPC patients(150 for training 50 for clinical promotion ), this study would use machine learning to 1.construct the relationship between dose absorbed and shape changes of parotid for prediction 2.design the strategy of dose constraint for accurately parotid sparing using the feedback coming from predicted geometric changes of the parotid. The implementation of the study is helpful to reduce the parotid dose ,maintain protid volume,and decrease the probability of xerostomia.