The incidence of oropharyngeal cancer has been steadily increasing during the past decades. This increase is linked with human papillomavirus, one of the most common sexually transmitted diseases in Canada and worldwide. Recent studies have shown the importance of using p16 testing to assess the HPV status of all oropharyngeal cancer patients on diagnostic. However, that practice was not common during early 2000, making historical data flawed. Many imputation models have been built to retroactively predict the HPV status of oropharyngeal cancer patients that were not tested. This models are based on clinical data, which is easy to store and analyze. However, recent advancements in the field of radiomics have enabled the use of CT scans obtained from patients to build models for cancer behavior. In this study, we take a novel approach to HPV status imputation by building machine learning models that utilize not only clinical data but also imaging features, aiming to show a significant improvement over classical models. The increase of performance between state of the art clinical models and our models will be assessed through the use of the RADCURE dataset from the Princess Margaret

Outgoing