A Knowledge-Based Planning Model for Proton-Based CSI
Abstract
Purpose
Leptomeningeal metastases (LM), or the spread of cancer to the cerebrospinal fluid, is a neurologically devastating manifestation of cancer. Without treatment survival is limited to months, and few effective therapies exist. A randomized phase II study recently showed that proton craniospinal irradiation (pCSI), more than doubles survival in patients with LM from breast and lung cancer. However, proton treatment planning is complex and time-consuming. Because LM typically progresses rapidly, fast treatment initiation and optimized planning workflows are needed. Here, we present a knowledge-based planning (KBP) model that automates plan optimization with the aim of increasing workflow efficiency.
Methods
We generated the KBP model using a cohort of 33 patients with LM who previously underwent pCSI to 30 Gy(RBE) in 10 fractions. KBP optimization objectives underwent several iterations of tuning until plan quality was considered adequate. The final model was tested with a cohort of 10 LM CSI patients. We recorded KBP plan optimization time and compared dose-volume histogram (DVH) parameters of KBP-developed and manual plans using paired t-tests (p<0.05). We employed robust optimization and normalized both manual and KBP plans to identical target coverage.
Results
Initial plan optimization required 20 minutes per plan without any need for manual intervention. A subsequent continued optimization reduced plan hot spots and required an additional 8 minutes. Comparing plan quality, KBP plans significantly increased CTV D1% (manual vs KBP: 104.7±0.2% [mean±standard deviation] vs. 105.8±1.1%). There was no significant difference in the CTV D99% between groups, and across 37 organ-at-risk parameters, differences were less than 1 Gy respectively.
Conclusion
We report a KBP model that reduces treatment planning time for proton CSI, requires minimal manual input, and achieves similar plan quality to standard manual planning. This may enable more rapid treatment initiation for patients with LM, and, by extension, improve patient outcomes.