Evaluating effectiveness of clustering algorithms in multiple target stereotactic radiosurgery.

Area of Science:

• Medical Physics
• Radiotherapy
• Computational Biology

Background:

• Single-isocenter-multiple-target stereotactic radiosurgery (SRS) reduces treatment time but risks dose coverage issues from rotational errors.
• Clustering targets can minimize isocenter-target distances, mitigating rotational uncertainty in SRS.

Purpose of the Study:

• To introduce a comprehensive SRS Target Clustering Framework (Framework) for evaluating clustering algorithms.
• To assess the effectiveness of different clustering algorithms in generating efficient SRS target configurations.

Main Methods:

• The Framework utilized agglomerative, weighted agglomerative, k-means, and weighted k-means clustering algorithms.
• Four optimization objectives were defined based on isocenter-target distance and its ratio to target radius.
• The Framework was applied to 126 SRS plans and compared against brute force ground truth solutions.

Main Results:

• Agglomerative clustering showed slightly higher average maximum isocenter-target distances (4.8 cm) compared to ground truth (4.6 cm).
• K-means and weighted k-means clustering demonstrated close agreement (within 0.1 precision) with ground truth for root-mean-square distances and ratios.
• The Framework effectively generated clusters for SRS targets, with k-means algorithms showing superior performance.

Conclusions:

• The SRS Target Clustering Framework is effective for optimizing SRS treatment planning.
• K-means and weighted k-means algorithms are highly accurate for minimizing uncertainty in SRS.
• This study is the first to investigate clustering algorithms for both minimax and sum-of-squares uncertainty in SRS.