Assessment of a diagnostic program for autodelineation of head and neck cancer based on PET/CT images

Abstract

Delineation of tumors and cancerous lymph nodes in medical imaging is a challenging, time-consuming and complex part of radiotherapy planning. A program for autodelineation of cervical cancer from MRI data was investigated to evaluate it’s possible use on PET/CT images. In this Master’s thesis an autodelineation program developed to identify cervical cancer tumors from different types of MR images was investigated. This program classifies every voxel in MR image stacks as either cancerous or non-cancerous, using voxel intensities, spatial relationships and Fisher’s Linear and Quadratic Discriminant Analysis (LDA and QDA).The aim of this thesis was to further develop the autodelineation program and adapt it to delineate head and neck cancers in PET/CT images. The dataset used in this study consisted of 206 head and neck cancer (HNC) patients who had undergone 18F-FDG PET and contrast-enhanced CT in conjunction with radiotherapy. All patients were treated at Oslo University Hospital (OUS), Norway between 31.10.2007 and 31.07.2015.