MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Development of a Hierarchical Clustering Algorithm to Aid in Automated Planning of Multiple Metastases Intracranial Stereotactic Radiosurgery Treatments

C Yeboah1*, M Ruschin1 , B Chugh1 , A Sarfehnia1 , A Sahgal1 , Y Lee1 , (1) Sunnybrook Health Sciences Centre and University of Toronto, Toronto,ON

Presentations

(Sunday, 7/14/2019) 2:00 PM - 3:00 PM

Room: Stars at Night Ballroom 4

Purpose: To develop and implement an algorithm for automated multi-target composite treatment plan split-up into multiple and practically-deliverable daily treatment plans for stereotactic radiosurgery (SRS) treatments.

Methods: A hierarchical clustering algorithm was developed and implemented clinically for generating multiple and stand-alone daily treatment plans from a single composite plan. The algorithm is based on concept of target clusters and their automatic identification. Two targets belong to the same cluster if their separation is less than their threshold separation. The latter is defined as the product of their average diameter plus 10mm and the target distribution-index (d-index) of the neighbourhood in which they reside. The d-index (0.5-20) is initialized such that, at first, it results in only one or few overcrowded/unstable clusters. Each overcrowded cluster is then recursively split-up into stable sub-clusters by iteratively decreasing its d-index and hence the threshold separations of target pairs in it. Finally, the desired number of daily plans is generated by utilizing established target clusters as building blocks. To test the algorithm, five experienced planners independently and manually split composite plans for five patients, each with 27-45 brain metastases, into 5-7 daily treatment plans. Observers-consensus cluster consists of targets in close-proximity grouped together by at least 4/5 observers. The developed algorithm’s agreement with observers-consensus clusters for each patient was assessed.

Results: Observers had consensus on 7-10 clusters for each of cases #1-5. The algorithm identified all observer-consensus clusters in 4/5 cases. In case #5, it identified 7/9 clusters. Overall, the algorithm identified 95% of observers-consensus clusters. The application took less than 2 minutes per patient to produce daily split-up plans whereas manual plan splitting approach took 30-90 minutes.

Conclusion: The proposed algorithm moves us a step closer towards a fully-automated planning workflow for multi-metastases SRS. It prevents errors such as target duplication and/or missed targets.

Keywords

Treatment Planning, Radiosurgery, Gamma Knife

Taxonomy

Not Applicable / None Entered.

Contact Email