基于三维预测剂量的自调节调强放疗自动计划方法

目的 开发一种基于预测剂量的自调节调强放射治疗自动计划方法，以增强自动计划的鲁棒性。方法 利用3D U-Res-Net_B网络预测出三维剂量分布后，在直接子野优化的每次迭代中先基于上次迭代结果计算当前剂量，再联合预测剂量计算目标剂量，然后以此为目标进行优化。完成所有迭代或满足循环退出条件后，得到最终的治疗计划。在30例直肠癌病例上进行测试，验证算法的效果。结果 临床计划治疗靶区的V_100%均值和标准差为（95.03±0.91）%，自动计划为（94.67±1.96）%，接近临床值（P>0.05），而预测值为（92.90±2.13）%，与临床计划的差异具有统计学意义（t=29.0，P<0.05）；自动计划在小肠V₃₅、膀胱V₄₀、股骨头的V₂₀~V₄₀等多项指标上低于预测值和临床值，且差异具有统计学意义（t=4.5~118.0，P<0.05），在其他危及器官的指标上与临床值的差异无统计学意义（P>0.05）。结论 本方法增强了自动计划的鲁棒性，提高了其应对复杂情况的能力。

Self-adjustable automatic planning method of intensity modulated radiotherapy based on 3D predicted dose

Objective To develope a self-adjustable automatic planning method of intensity modulated radiotherapy based on predicted dose, in order to enhance the robustness of automatic planning. Methods After the patients'' dose by 3D U-Res-Net_B network was predicted, the current dose was calculated based on the last iteration result, then the predicted dose was combined to calculate the target dose and optimized. With all iterations completed or exit conditions satisfied, final treatment plannings would be acquired. A total of 30 cases of rectal cancer were tested to verify the effectiveness of the algorithm. Results The mean value of planning target volumes'' V_100% was (95.03±0.91)% for clinical plans, close to (94.67±1.96)% for automatical plans(P>0.05), and better than (92.90±2.13)% for predicted dose with the statisically significant difference (t=29.0,P<0.05). Automatic planning''s indexes such as V₃₅ of small intestines, V₄₀ of bladders and V₂₀-V₄₀ of femoral heads were lower than predicted and clinical ones, with the statisically significant difference(t=4.5-118.0, P<0.05). Discrepancy in other indexes of organs at risk was not statistically significantly different(P>0.05). Conclusions This method made automatic planning processes more robust and more adaptive to difficult clinical situations.