Aim of the PhD Project:
- Design and optimisation of AI tools suitable for segmentation of multiple organs.
- Co-registration and feature learning from both CT and PET data within these tools.
- Combining the above into one multi-organ segmentation tool for assessing tumour and healthy tissue metabolism.
- Test and validate in clinical data sets
Project Description / Background:
This project aims to establish a new paradigm that will enable prediction of patient response, side-effects and survival following specific anti-cancer treatments from PET-CT imaging data. State-of-the-art AI methods will be designed to perform a comprehensive analysis of whole-body PET-CT images to extract information not currently available or utilized in large scale data-analysis or in clinical routine. A key aspect of this project is the availability of a series of standardized datasets of PET-CT scans from patients taking part in phase II/III clinical trials curated by the UK PET Core Lab at St Thomas’ Hospital.
Imaging plays a pivotal role in diagnosis, response evaluation and surveillance of patients with cancer. PET-CT with the tracer FDG has since its introduction two decades ago been the most rapidly growing imaging technology. The basis for imaging with FDG PET-CT is accelerated glucose metabolism in cancer cells. However, this is not specific for cancer and FDG is readily taken up by other cells e.g. activated lymphocytes, neutrophils and macrophages, as well as in immune related healthy tissue such as spleen and bone marrow. This lack of specificity is considered a major drawback with regard to the clinical use of FDG-PET in patients with cancer. However, in recent years our improved understanding of the complex role of the immune system for the development of cancer, makes it increasingly relevant to examine this metabolic overlap. In this scenario FDG-PET provides a unique possibility for non-invasive, simultaneous whole-body monitoring of changes in the metabolism of tumour as well as in immune related healthy tissue including the gut. Currently, information on healthy tissue is not systematically evaluated nor utilised, mainly because extraction of this information on a larger scale is extremely labour intensive and associated with a number of technical challenges.
Applying AI in this scenario would potentially enable an automated faster and more robust way of extracting this information from a large number of PET-CT images, leading to improved evaluation of healthy tissue metabolism as a biomarker for use in clinical trials. Prediction based on comprehensive metabolic information from PET imaging would potentially enable us to predict adverse events; e.g. susceptibility of lung tissue to radiotherapy, or cardiac tissue to chemotherapy. It might also qualify very early evaluation and prediction of therapy response by combining information of changes in tumour metabolism with changes in immune related healthy tissue and bowel during therapy. This will require developing, validating and applying new methods for automated multi-organ segmentation of PET/CT and tumour characterisation using deep learning approaches based on convolutional neural networks and other deep network architectures.The PhD-candidate should preferably have a background in computer science, physics or biomedical engineering, preferably with some experience in (medical) image analysis and/or deep learning. You will acquire in-depth training in all aspects of PET-CT scanning, especially image reconstruction and analysis, but also image analysis, deep learning, and clinical application. This project will also involve close collaboration with the newly established London Medical Imaging and AI Centre for Value-Based Healthcare based at St Thomas’ .