A Fully Convolutional Normalization Approach of Head and Neck Cancer Outcome Prediction

William Le, Francisco Perdigón Romero, Samuel Kadoury

Abstract: Medical image classification performance worsens in multi-domain datasets, caused by radiological image differences across institutions, scanner manufacturer, model and operator. Deep learning is well-suited for learning image features with priors encoded as constraints during the training process. In this work, we apply a ResNeXt classification network augmented with an FCN preprocessor subnetwork to a public TCIA head and neck cancer dataset. The training goal is survival prediction of radiotherapy cases based on pre-treatment FDG-PET/CT scans, acquired across 4 different hospitals. We show that the preprocessor sub-network acts as a embedding normalizer and improves over state-of-the-art results of 70% AUC to 76%.

Paper Type: well-validated application

TL;DR: UNet subnetwork for domain adaptation in medical image classification

Track: short paper

Keywords: Classification, head and neck cancer, deep learning, PET-CT, UNet-FCN, multi-domain, radiotherapy, outcome survival prediction

Presentation Upload Agreement: I agree that my presentation material (videos and slides) will be made public.