Institut de Neuroscience des Systèmes, Functional Neurosurgery and Radiosurgery Department, Aix-Marseille Université, Hôpital de la Timone, APHM, Marseille, France.
Service de Neurochirurgie Fonctionnelle & Radiochirurgie, Hôpital d'adulte de la Timone, 264 Bvd Saint Pierre, 13 285, Marseille Cedex 05, France.
Acta Neurochir (Wien). 2021 Apr;163(4):981-989. doi: 10.1007/s00701-020-04695-x. Epub 2021 Jan 5.
Gamma Knife radiosurgery (GKRS) inverse dose planning is currently far from competing effectively with the quality of dose planning developed by experienced experts. A new inverse planning (IP) method based on « efficient convex optimization algorithms » is proposed, providing high-quality dose plans in real time.
Eighty-six patients treated by GKRS for vestibular schwannomas (VS) were recruited. The treatment plans created by the first author, who has 27 years of experience and has developed and delivered more than 15,000 dose plans, served as reference. A first set of basic constraints determined by default led the IP for an initial real-time dose plan. Additional constraints were interactively proposed by the planner to take other parameters into account. A second optimized plan was then calculated by the IP. The primary endpoint was the Paddick Conformity Index (PCI). The statistical analysis was planned on a non-inferiority trial design. Coverage, selectivity, and gradient indexes, dose at the organ(s) at risk, and 12 Gy isodose line volume were compared.
After a single run of the IP, the PCI was shown to be non-inferior to that of the "expert." For the expert and the IP, respectively, the median coverage index was 0.99 and 0.98, the median selectivity index 0.92 and 0.90, the median gradient index 2.95 and 2.84, the median dose at the modiolus of the cochlea 2.83 Gy and 2.86 Gy, the median number of shots 14.31 and 24.13, and the median beam-on time 46.20 min and 26.77 min. In a few specific cases, advanced tools of the IP were used to generate a second run by adding new constraints either globally (for higher selectivity) or locally, in order to increase or decrease these constraints focally.
These preliminary results showed that this new IP method based on « efficient convex optimization algorithms », called IntuitivePlan®, provided high-quality dose plans in real time with excellent coverage, selectivity, and gradient indexes with optimized beam-on time. If the new IP evaluated here is able to compete in real time with the quality of the treatment plans of an expert with extensive radiosurgical experience, this could allow new planners/radiosurgeons with limited or no experience to immediately provide patients with high-quality GKRS for benign and malignant lesions.
伽玛刀放射外科（GKRS）的逆向剂量规划目前远不能有效地与经验丰富的专家所开发的剂量规划质量相媲美。本文提出了一种新的基于“高效凸优化算法”的逆向规划（IP）方法，能够实时提供高质量的剂量计划。
本研究共纳入了 86 例因前庭神经鞘瘤（VS）接受 GKRS 治疗的患者。本研究以具有 27 年经验并开发和交付了超过 15000 个剂量计划的第一作者所创建的治疗计划作为参考。首先，由默认设置确定的一组基本约束条件引导 IP 生成初始实时剂量计划。然后，规划师可通过交互方式提出附加约束条件，以考虑其他参数。然后，通过 IP 计算第二个优化计划。主要终点是 Paddick 适形指数（PCI）。该研究计划进行非劣效性试验设计的统计学分析。比较了覆盖指数、选择性指数、梯度指数、危及器官剂量和 12Gy 等剂量曲线体积。
单次运行 IP 后，其 PCI 被证明不劣于“专家”的计划。对于专家和 IP 生成的计划，其覆盖指数中位数分别为 0.99 和 0.98，选择性指数中位数分别为 0.92 和 0.90，梯度指数中位数分别为 2.95 和 2.84，耳蜗蜗轴处的剂量中位数分别为 2.83Gy 和 2.86Gy，照射野数中位数分别为 14.31 和 24.13，以及治疗时间中位数分别为 46.20min 和 26.77min。在少数特定情况下，通过添加新的全局（用于提高选择性）或局部约束条件（用于增加或减少局部约束条件），使用 IP 的高级工具生成第二次运行。
这些初步结果表明，这种基于“高效凸优化算法”的新型 IP 方法，称为 IntuitivePlan®，能够实时提供高质量的剂量计划，具有优异的覆盖指数、选择性指数和梯度指数，同时优化了治疗时间。如果本文中评估的新 IP 能够实时与具有丰富放射外科经验的专家治疗计划的质量相媲美，那么这将使经验有限或缺乏经验的新规划师/放射外科医生能够立即为良性和恶性病变患者提供高质量的 GKRS。
