RADI-15. CLUSTERING AND GROUPING OF BRAIN METS IN RADIOSURGERY TREATMENTS. (12th August 2019)

• Record Type: Journal Article
• Authors: Salah, Khaled; Mckenna, John; Jozsef, Gabor; Knisely, Jonathan
• Is Part Of: Neuro-oncology advances. Volume 1(2019)Supplement 1
• Page Start: i24
• Page End: i24
• Publication Date: 2019-08-12
• Subjects: 616.99481
• DOI: 10.1093/noajnl/vdz014.108
• Languages: English
• ISSNs: 2632-2498

Abstract

Abstract: INTRODUCTION: Radiosurgical treatment of numerous lesions in the brain with 'single-isocenter' radiosurgery on a linac often requires using multiple isocenters. With our TPS (Elements, Brainlab) multiple plans need to be generated for each set of lesions, and a sum plan calculated. We investigated how to distribute multiple lesions into two groups for two isocenters to achieve a good summed dose distribution. METHODS: The DICOM RS file is exported and the PTV data is extracted by a MATLAB program that calculates the convex hulls, estimated radii, and the centers of mass for each PTV. Two approaches were tried: (1) Lesions close to each other (closer than a certain limit) are put in different groups and (2) Create clusters by kMeans clustering, which allows close lesions but the groups are distant from each other. MATLAB programs were written for all approaches. Treatment plans were generated for three patients (20, 13, 15 lesions) using each method and compared with the actual treatment plan used to treat the patient based on the intuitive grouping of lesions by the planners. Dose maximums outside the lesions, and volumes in the normal tissue exceeding 75, 50 and 25% of the prescription dose were evaluated. RESULTS AND DISCUSSION: The coverage of all lesions for all plans were 95% of the prescription dose. The first approach allowed lowering the maximum dose between lesions, but with summing dose distributions this advantage disappeared. The maximum dose and the 75, 50 and 25% dose volumes were also all worse than in plans generated by experienced planners and higher normal brain doses are delivered if closely spaced lesions are separated into different isocenters for treatment. However, the clustering approach resulted in the same or better values of these same parameters, i.e. improved dose distributions over the dosimetrist's intuitively chosen separation.