Deep Learning Based HPV Status Prediction for Oropharyngeal Cancer Patients

Infection with the human papillomavirus (HPV) has been identified as a major risk factor for oropharyngeal cancer (OPC). HPV-related OPCs have shown to be more radiosensitive and reduced related death. Hence, histological determination of HPV status patients depicts an essential diagnostic factor. We investigated ability deep learning models imaging based detection. To overcome problem small medical datasets, we used transfer approach. A 3D convolutional network pre-trained on sports video clips was fine-tuned, such that full information in CT images could exploited. The model able differentiate HPV-positive from HPV-negative cases, area under receiver operating characteristic curve (AUC) 0.81 external test set. In comparison neural (CNN) trained scratch 2D architecture ImageNet, performed best. Deep are capable image-based determination. Video pre-training improve training data, but further studies needed verification.