VIBRANT: Targeting Variations in gastroIntestinal cancer Biology with MRI and radiotherapy using Artificial iNTelligence

This project will focus on gastrointestinal cancers (pancreatic and colorectal) that often have poorer outcomes after standard treatment. This project will use ultra-high field MRI to characterise tumour heterogeneity at microscopic resolutions and correlate MRI findings with ‘ground-truth’ histopathology. Deep learning (artificial intelligence) will be used for the first time in cancer to characterise cancer heterogeneity with MRI, and translate these findings to clinical MRI to enhance the quality and resolution of clinical MRI images. MRI whole tumour ‘virtual biopsy’ will provide previously unobtainable information on tumour microenvironment and biologic behaviour (e.g. oxygenation, metabolism, cellularity and organisation). This will enable personalisation of patient treatment and allow MRI-Linacs and other therapies to target changes in cancer biology, leading to better cancer control and less side-effects. Deep learning models that will be explored include (i) deep learning super-resolution - a method of obtaining a higher resolution image from clinical MR image (ii) convolutional neural network (iii) generative adversarial network. Radiomics analysis (computer-aided high-throughput analysis of medical images that can extract hundreds to thousands of quantitative or textural imaging parameters) may also be used in the analysis of ex vivo and clinical MRIs. The ability of developed deep learning models to produce a ‘high-resolution’ clinical MRI (i.e. whole tumour MRI ‘virtual biopsy’) will be tested on patient clinical MRI images. Patient involvement in this study includes an additional MRI performed anytime between diagnosis and surgery, and consent to tissue donation/access to their surgical specimen for MRI scanning. The MRI scan will be performed by suitably qualified allied health professionals, using a scan protocol specifically developed for the study to ensure adherence to the study intervention. The duration of the scan is approximately 30-45mins, and may include the administration of a contrast agent. MRI acquisition: Standard clinical multiparametric MRI examination including anatomical (T2-weighted) and Diffusion Weighted Imaging (DWI) will be acquired. Research functional sequences will also be acquired. Clinical imaging of gastrointestinal tumours is affected by respiratory and bowel motion, and the sequences will be optimised to either short acquisition times, or respiratory gating within the time period of a standard clinical examination. A range of in vivo DWI datasets with different b-values and directions will be acquired. Various models will be applied (e.g. non-gaussian) to gauge heterogeneity in tumour cellularity, vasculature and organisation. Dynamic Contrast Enhanced (DCE) MRI will also be acquired with an optimised temporal and spatial resolution and perfusional models constructed (e.g. Tofts). Other sequences will be acquired to assess tumour heterogeneity: these include advanced diffusion sequences e.g. Diffusion Tensor Imaging (DTI), Intravoxel Incoherent Motion, (IVIM) to assess cellularity, Chemical Exchange Saturation Transfer (CEST) to assess metabolism and Blood Oxygenation Level Dependent (BOLD) MRI and Mapping of Oxygen By Imaging Lipids relaxation Enhancement (MOBILE) will be acquired to assess tumour oxygenation. Works-in-progress sequences may be used. These clinical images will be compiled into an atlas for clinical validation in this project and future studies. Patients will undergo standard treatment as recommended by their treating team. Patients will undergo standard surgery +/- radiotherapy +/- chemotherapy +/- molecular targeted therapies +/- immunotherapy. There is no change to patient treatment on this study.