Head and Neck Cancer

PV QA 2 - Poster Viewing Q&A 2

MO_33_2763 - A Radiomics Signature for Treatment Stratification in Advanced and Recurrent Nasopharynx Cancer

Monday, October 22
10:45 AM - 12:15 PM
Location: Innovation Hub, Exhibit Hall 3

A Radiomics Signature for Treatment Stratification in Advanced and Recurrent Nasopharynx Cancer
G. Kusumawidjaja, Y. Li, S. A. Gan, J. H. Phua, Y. Y. Ng, J. T. Wee, K. W. Fong, T. W. K. Tan, Y. L. Soong, K. Sommat, F. Q. Wang, and M. L. K. Chua; National Cancer Centre Singapore, Singapore, Singapore

Purpose/Objective(s): Viral-associated head and neck cancers are associated with distinct radiological features that may predict tumor biology. Radiomics involves multi-feature analyses of voxel intensity, tumor shape and texture, and may offer additional information on the tumor biological architecture, complementing clinical indices in predicting tumor behavior and prognosis. Here, we investigate a 104-feature radiomics algorithm in predicting outcomes in locally advanced and locoregionally recurrent EBV+ nasopharynx cancer (NPC).

Materials/Methods: We utilized a cohort of 225 NPC cases (107 treatment-naïve locally advanced [LA-NPC]; 118 locally recurrent cases [LR-NPC]), treated at a single academic institution from 2002 - 2014. Gross tumor volume (GTV) was segmented on a CT dataset. Radiomics features were extracted using Pyradiomics v1.3.0. Primary clinical endpoint was overall survival (OS).

Results: Multivariable analyses revealed that age, T-cat, GTV were significant predictors of OS in the LA-NPC and LR-NPC cohorts; clinical model alone offered a Harrell’s C index of 0.61 and 0.71 for OS, respectively. Of the 104 indices, 14 were significantly associated with GTV (R > .750, p < 0.001); 7 (firstorder: energy, range, total energy; glszm: zone entropy; ngtdm: coarseness; shape: max 2D diameter row, surface vol ratio; p < 0.0005) were enriched in the LR-NPC compared to the LA-NPC cohort, suggesting that they may be associated with radioresistance. Next, these enriched features were significantly associated with OS in the LR-NPC cohort on multivariable analyses; the radiomics-alone signature provided a C index of 0.72 for OS. Importantly, a combinatorial clinico-radiomics model strongly predicted for 3-y and 5-y OS in the LA-NPC (AUC = 0.803, 3-y; 0.756, 5-y) and LR-NPC (AUC = 0.814, 3-y; 0.791, 5-y) cohorts.

Conclusion: Radiomics-extracted features have the potential to predict for therapeutic resistance in EBV+ NPC. Here, we developed a radiomics signature that complements conventional clinical indices in OS prediction.

Author Disclosure: G. Kusumawidjaja: None. Y. Li: None. S. Gan: None. J. Phua: None. T.W. Tan: None. K. Sommat: None.

Grace Kusumawidjaja, MD

Presentation(s):

Send Email for Grace Kusumawidjaja


Assets

MO_33_2763 - A Radiomics Signature for Treatment Stratification in Advanced and Recurrent Nasopharynx Cancer



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for A Radiomics Signature for Treatment Stratification in Advanced and Recurrent Nasopharynx Cancer