Author: Josephine Chen, CheukKai Becket Hui, Yildirim D. Mutaf 👨🔬
Affiliation: Kaiser Permanente 🌍
Purpose:
To demonstrate the effectiveness of a target clustering program in generating cluster configurations and isocenter placements for multiple brain lesions in SRS treatment planning, with the aim of enhancing overall treatment accuracy.
Methods:
We developed a target clustering program to address rotational uncertainties in multiple-target SRS treatments, balancing efficiency and accuracy. The program, implemented on the MIM interface, applies complete-linkage agglomerative clustering to create clusters and utilizes the minimum sphere algorithm to compute isocenters. This approach aims to minimize the maximum shift-rotation coefficient, which quantifies the potential dose coverage displacement due to rotational errors. Following the computations, the program generates optimized planning and isocenter structures based on the clustering results. To validate its effectiveness, we retrospectively analyzed clinical SRS courses containing 10 or more targets. These courses, where multiple clusters were previously formed heuristically, allowed for comparison between the program's output and existing clinical practices.
Results:
Our analysis identified 29 eligible SRS courses (17 with 2 clusters and 12 with 3 clusters), with an average of 15.7 PTVs per course and a maximum distance between targets of 13.7 cm. The program demonstrated remarkable efficiency, completing clustering tasks in approximately 10 seconds. In 22 of the 29 cases, the program produced a lower maximum shift-rotation coefficient compared to traditional approaches. Statistical analysis showed a significant reduction in the average maximum shift-rotation coefficient from 0.97 mm/° in clinical practice to 0.87 mm/° using the program (p = 0.002). Notably, in 4 out of the 12 courses treated with 3 clusters, the program's 2-cluster configuration achieved a smaller maximum shift-rotation coefficient.
Conclusion:
The proposed clustering program shows promising potential in generating effective cluster configurations for multiple-target SRS. By reducing processing time and mitigating rotational uncertainty, the program could become a valuable tool in SRS treatment planning.