MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Radiomics Features Predict Tumor Volume Change During Head and Neck Cancer Radiotherapy

M Surucu1*, I Mescioglu2 , F Cozzi1 , N Hurst1 , B Emami1 , J Roeske1 , (1) Loyola University Medical Center, Maywood, IL, (2) Lewis University, Romeoville, IL,

Presentations

(Tuesday, 7/31/2018) 7:30 AM - 9:30 AM

Room: Davidson Ballroom B

Purpose: To identify radiomics parameters that predict the change in the gross tumor volumes of head and neck (HN) patients during radiation therapy using decision tree analysis.

Methods: Forty HN patients that were rescanned and replanned during radiotherapy were retrospectively analyzed. Primary gross tumor volumes (GTVp) on the initial and rescan CTs were contoured by radiation oncologists. The percentage change in primary gross tumor volumes ΔGTVp was calculated for each patient. The GTVp that were on the initial CT scan were used to calculate radiomics features using the IBEX software. A total of 1020 features were calculated. A decision tree analyses was performed to identify the features that predicts the ΔGTVp, where ΔGTVp was taken as a continuous parameter. A univariate density regression based decision tree induction algorithm was implemented.

Results: Median dose at rescan was 39.6Gy and median ΔGTVp was 29.3%. The decision tree induction software identified 5 groups of ΔGTVp including: tumor growth; tumor shrinkage groups of ≤21%, 22-43%, 44-65%, and >65%. The final decision tree consisted of four levels, and used 7 of the radiomics features and predicted whether the patient will be placed in one of the 5 ΔGTVp groups. The radiomics features used were 4-7Energy, 11-7Correlation, 11-4Contrast, 0ShortRunLowGrayLevelEmpha, 10-4ClusterProminence, 0ShortRunHighGrayLevelEmpha, and 2-7Correlation. The accuracy of the decision tree was 95%, with only two patients being marginally misplaced among ΔGTVp groups, belonging to the next group up/down.

Conclusion: The information hidden in the CT images was extracted by the radiomics analysis and helped to stratify patients according to the change in their primary tumor volume. Since this information is available at the beginning of the treatment, flagging patients that will undergo higher levels of tumor shrinkage can help place these patients into the adaptive care paths resulting in judicious use of the limited clinical resources.

Funding Support, Disclosures, and Conflict of Interest: Murat Surucu and John Roeske have research grant from Varian Medical Systems and NIH, not related to this work.

Keywords

CT, Image Analysis, Tumor Control

Taxonomy

IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics

Contact Email