External validation of a radiomic signature to predict HPV (p16) status from standard CT images of anal and vulvar cancer patients.

Abstract

e15502 Background: HPV status of anal and vulvar cancers cannot be predicted by visual inspection as well as for oropharyngeal cancers. Radiomics applied on computed tomography images can extract features that may better characterize the structure and the underlying biology of the tumor. Methods: In this multi-center study, we validated a CT based radiomic signature to predict HPV (p16) status, developed in head & neck cancer, in anal and vulvar cancer patients. The patients cohort was composed of 68 anal cancer patients and 7 vulvar cancer patients, with p16 status determined by immunohistochemistry, while a control cohort was composed of 422 lung cancer patients. The patient cohorts come from 4 different centers (Maastro Clinic - the Netherlands, CHU Liege – Belgium, St Luke’s Hospital – Ireland, Cork University Hospital - Ireland). The primary tumor volume was manually delineated for each patient on axial CT images. Prior to analysis, all images were resampled to isotropic voxels of 2 mm, using linear interpolation. A total of 37 radiomics features were calculated from five groups: tumor intensity, shape, texture, Wavelet and Laplacian of Gaussian. The signature was built using regularized logistic regression [1]. The signature was evaluated according to the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) and the Radiomics Quality Score (RQS). Results: The signature classified anal and vulvar cancers based on their HPV status (positive or negative), with an AUROC of 0.760 comparable to the performance of the original signature developed in oropharyngeal squamous cell carcinomas (AUROC of 0.764) [1]. The model, tested in the control cohort of lung cancer patients, predicted the HPV positive status of 1% of the patients which is in line with expected European prevalence (0 – 10%). This signature is TRIPOD level 4 (57%) with an RQS of 61%. Conclusions: This study supplies an additional insight into HPV imaging phenotype, providing a proof of concept that molecular information can be inferred from standard medical images by means of radiomics. These preliminary but encouraging results may pave the road for further generalization of CT image features of HPV-related tumors and aid in the optimization of future therapy developments [2]. Reference [1] Ralph TH Leijenaar et\xa0al., The British Journal of Radiology 2018 91:1086 [2] Immunotherapy Drug with Two Targets Shows Promise against HPV-Related Cancers - accessed on 12/02/2021 - https://www.cancer.gov/