Extracting greater data from standard CT imaging in non-small cell lung cancer.

Abstract

Lung cancer is the leading cause of cancer death worldwide. Patients with lung cancer have a very poor outcome compared to other common cancers. The aim of this project is to identify whether CT data accumulated by patients with Non-Small Cell Lung Cancer (NSCLC) provide an untapped resource that can improve their quality of care, by generating more information from existing imaging. Extracting additional data from imaging is termed radiomics. One way of extracting this data is using mathematical descriptions of an image or regions of interest within an image. This type of assessment is termed Textural Analysis (TA). The experimental work described in this thesis uses a second order TA technique to analyse CT (Computer Tomography) imaging data from patients with NSCLC. The software developed as part of this project uses a voxel by voxel analysis of CT imaging, by comparing grey levels using grey level co-occurrence matrices (GLCMs). Three approaches (themes 1-3) were explored. The first approach shows that TA can help differentiate between tumour and Radiation Induced Lung Injury (RILI) after Stereotactic Ablative Body Radiotherapy (SABR) in early stage lung cancer. The second approach suggests that assessing muscle loss on diagnostic imaging can help predict outcome in advanced NSCLC. The third and final approach uses TA to generate a functional assessment of lung function. TA is able to differentiate between patients who are fit or unfit for radical radiotherapy, based on TA of lung tissue on CT imaging, rather than formal lung function tests. TA technique described in this thesis is a novel intervention in gaining functional data from CT imaging. It is particularly attractive as the analysis is generated from routine oncological imaging. As a result these tools have the potential to be cost effective and could be integrated into a standard radiology workflow.