(Tuesday, 7/16/2019) 10:00 AM - 10:30 AM
Room: Exhibit Hall | Forum 6
Purpose: To identify tumor and patient characteristics that can predict the absolute volume change for small and large primary and nodal tumors of the head and neck region.
Methods: Ninety-two patients that received chemo-radiotherapy andrescanned during treatment were retrospectively identified. The primary (GTVp) and nodal (GTVn) tumors were contoured on both initial and rescan CTs, and the change in volume was calculated. For each primary and nodal initial tumor volumes, the patients were divided into two size-based groups (small and large) based on the initial median volumes. Fourteen clinical and tumor parameters were identified including age, gender, site, chemotherapy, Karnofsky performance status, HPV status, tumor growth pattern, tumor grade, staging (tumor, nodal and group), initial volumes of GTVp, GTVn and GTVtotal. The C4.5 decision tree induction algorithm was used to construct decision trees.
Results: Median initial GTVp and GTVn were 29.1 cc (range: 2.6 to 110.8) and 21.6 cc (0.8 to 495.7), respectively. The median change in absolute volume was 4.8 cc (-20.3 to 110.8) and 5.3 cc (-17.7 to 188.8) for GTVp and GTVn. The accuracy of the primary tumor decision tree was 80.6% where GTVp volume was the main deciding branch followed by tumor stage, HPV status, group stage, site, age, GTVn, growth pattern, chemotherapy and site which were found to be correlated with change in absolute GTVp. For the nodal decision tree, the accuracy was 85.7% and after the GTVn main branch, the tumor stage, tumor grade, age, GTVtotal, site, and chemotherapy were identified to be correlated with the absolute GTVn change.
Conclusion: Two decision trees were constructed that successfully relate the patient/tumor characteristics with the change in absolute volume for both primary and nodal tumor volumes. This tool can possibly be used to identify patients in need of adaptive radiotherapy before they start treatment.
Funding Support, Disclosures, and Conflict of Interest: Murat Surucu and John Roeske have research grant support from NIH and Varian Medical Systems, not related to this study.
Keywords
Radiation Therapy, Image-guided Therapy, Tumor Control
Taxonomy
TH- response assessment : Machine learning
Contact Email
