Back to Projects

AI-enabled Imaging, Emerging Imaging

Biology-encoded magnetic resonance imaging and deep learning for precision radiotherapy planning in rectal cancer

Project ID: 2020_018

1st Supervisor: Isabel Dregely, King’s College London
2nd Supervisor: Marc Modat, King’s College London
Clinical Champions: Vicky Goh and Davide Prezzi, King’s College London

Aim of the PhD Project:

Hypoxia is a cause for radiotherapy resistance and present in up to 60% of cancers. We hypothesize that advanced imaging can detect hypoxia to improve radiotherapy outcomes. This project aims to:

  • Develop an MRI method to quantify oxygen and perfusion
  • Develop deep learning for radiotherapy plans based on biological targeting

Project Description / Background:

Hypoxia is a cause for radiotherapy resistance and is present in 60% of locally advanced cancers. The future of precision radiotherapy (including dose escalation & dose painting) relies on our ability to accurately identify radio-resistant cancer sub-volumes. We have preliminary data which show that using oxygen enhanced MRI with T1-/ T2*- and DCE-mapping sequences are able to detect hypoxia-related signal changes in colorectal tumours when giving 100% oxygen breathing (1). However, to enable precision, i.e. ‘biology-informed’ radiotherapy planning, the accuracy and robustness of hypoxia mapping has to be improved significantly. Technical challenges need to be overcome, as the effect is small and highly sensitive to organ motion. Importantly there is a complex biological link between hypoxia and perfusion, which means both parameters need to be quantified simultaneously, not in separate sequences to extract meaningful tumour hypoxia information.

We have recently developed a suitable acquisition scheme using magnetization-prepared segmented gradient echo imaging sequence, which also includes motion navigators and undersampling to improve efficiency and robustness for body imaging application (2,3). Here we propose to use a multi-contrast prepared sequence to encode perfusion and oxygen sensitive signals. We will use advanced reconstruction methods using simulated signal evolution dictionaries and including deep learning algorithms to extract hypoxia vs perfused tissue information from the biology-encoded images.

The project will be ideal for an engineering, physics/mathematics or computer science student who will be exposed to real-world clinical problems, multimodal medical imaging research & platform development. Close industry collaboration will enable the student to gain experience in one of the major medical imaging companies in the world and through clinical collaborators access to an established cancer imaging multicentre network for future multicentre clinical translation. 

References:

  1. Prezzi et al, abstract with oral presentation at ESGAR 2019
  2. Vidya Shankar, R. et al. Accelerated 3D T 2 w‐imaging of the prostate with 1‐millimeter isotropic resolution in less than 3 minutes. Magn. Reson. Med. 82, 721–731 (2019).
  3. Roccia, E. et al. Accelerated 3D T 2 mapping with dictionary‐based matching for prostate imaging. Magn. Reson. Med. 81, 1795–1805 (2019).

Back to Projects