Develop a framework to integrate lung CT scan with metabolomics data for patients with lung cancer and identify potential biomarkers

Lead Research Organisation: Imperial College London
Department Name: Dept of Surgery and Cancer

Abstract

The theme of my project deals with the application of novel quantitative/computational AI-based technologies in the area of CT imaging for pulmonary disease. As biomedical research becomes more mathematical and computing-dependent, the newly emerged field of radiomics is rapidly growing as a real game-changer making the medical image analysis faster and more efficient. This quantitative field is still in its infancy and lacks powerful automated frameworks that could be widely deployed in clinical practice to improve patients outcome. Specifically, in our project, we intend to analyse imaging data from a cohort of lung cancer patients.

As CT scans, currently are the most widely used imaging modality in lung cancer analysis, our research will be focusing on the quantitative modelling and processing of unstructured CT information to create clinically meaningful lung imaging features. Hopefully, they will allow us to obtain potentially important knowledge regarding the early detection and staging of the disease. Additionally, we would seek to understand whether patients prior pulmonary disorders increase the chances of lung cancer occurrence. This question will be tested by introducing prior lung disease information into the analysis. The results will be compared against the model that did not contain information on prior lung disease of the patient. Finally, we intend to use this radiomics analysis for further investigation with biomolecular data. Synthesizing the radiomics results with other omics data, most likely metabolomics, will hopefully lead us closer to the discovery of novel non-invasive biomarkers giving oncologists and radiologists more tools to improve patients outcomes.

The project computational analysis of the data will involve implementing both traditional statistical and machine learning methods as well as deep learning-based approaches and will consider using major existing platforms to perform lung segmentation function. However, feature extraction, their further processing, and possible synthesis with other omics data will have to be designed during our work and automated.

Publications

10 25 50