Author: Strahinja Stojadinovic, Robert Timmerman, Yulong Yan 👨🔬
Affiliation: Department of Radiation Oncology, UT Southwestern Medical Center, The University of Texas Southwestern Medical Center, University of Texas Southwestern Medical Center 🌍
Purpose: Radiosurgery for large (>10cc) arteriovenous malformations (AVMs) poses significant challenges due to increased risks of complications and lower obliteration rates. To mitigate toxicity, large AVMs are typically treated in multiple stages over time. Currently, the Leksell Gamma Plan (LGP) software lacks a dedicated clinical workflow for planning staged AVM treatments. Manual planning for such cases is intricate, time-consuming, and prone to error. This study presents an automated solution for staged AVM treatments using GammaKnife units.
Methods: At our institution, treatment planning for large AVMs typically involves prescribing a 20 Gy dose to the target periphery. This is achieved through manual LGP placement of multiple shots, all utilizing the same collimator size and weight. The generated LGP plan is subsequently exported to DICOManTX, an in-house software platform. Based on the user-defined number of stages, a conditioned K-means clustering algorithm is employed to group the shots into corresponding clusters. This process facilitates the automated generation of a staging report. The performance of this automated staging approach was evaluated by comparing it to manually created staging plans. The comparative analysis encompassed metrics such as average distance to cluster centroids, spatial variance within each stage, and variance in treatment time across stages.
Results: Average distance to centroids and spatial variance are two of four metrics used to measure cluster compactness. These metrics are comparable to those of manual plans, as is treatment time variance. However, the proposed method drastically reduces manual staging time from 1-2 hours to a few seconds.
Conclusion: The K-means clustering algorithm demonstrates significant potential for optimizing staged treatments of large AVMs within the GammaKnife framework. By eliminating the need to pre-define stage volumes, this approach offers enhanced flexibility in treatment planning while simultaneously achieving a substantial reduction in staging time.