Back to Projects

AI-enabled Imaging, Emerging Imaging

Simultaneous oxygen and perfusion magnetic resonance imaging (MRI) and deep learning for precision radiotherapy planning in rectal cancer

Project ID: 2019_064

1st supervisor: Isabel Dregely, King’s College London
Joint 1st supervisor: Vicky Goh, King’s College London

MRI is emerging as a tool for image-guided radiotherapy (RT) planning providing valuable anatomical and functional information. Hypoxia is a cause for radiotherapy resistance. MRI using T1-mapping (TOLD) or T2*-mapping (BOLD) sequences have been proposed to assess tumour oxygenation, which may enable personalised radiotherapy planning, i.e. dose adaption based on the underlying tumour biology. Measuring quantitative R1 = 1/T1 with TOLD, before and during 100% oxygen breathing, allows ∆R1(t) = R1 O2 (t) – R1air(t) extraction, which has been shown previously in an animal model to be related to hypoxia (1). Alternatively, the BOLD effect can be used to extract oxygen information based on ∆R2*. We and others have shown recently that using clinically available MR sequences, we are able to detect signal changes in colorectal cancer when breathing 100% oxygen.

Our preliminary data showed that while it was feasible to observe an effect (see figure), interpretation remains difficult; perfusion (delivery) and oxygenation are closely linked. Currently, these are acquired in separate scans and analysed independently. In our preliminary study, combined R1, R2* and perfusion quantification were unsuccessful in one third of the patients, due to bulk tumour motion. We observed significant ΔR1 signal changes in a subset of patients reflecting hypoxia severity, which correlated overall with low perfusion metrics. While this indicates that it is feasible to detect hypoxia using R1, R2* and perfusion measurements, an advanced MRI acquisition method that simultaneously assesses these parameters in a robust and efficient way would be a major step forward. Further, the method needs to accommodate RT-specific requirements: rapid scanning, as immobilisation for RT can be uncomfortable; good SNR efficiency as receiver coils are placed in the RT-setup, i.e. at further distance from body, full FOV, 3D and highly geometrically accurate imaging to produce pseudo CT for accurate dosimetry, and to delineate gross target volumes (GTV) and organs at risk (OAR) for radiotherapy plans.

We have recently developed a suitable acquisition scheme using T2-prepared and very recently diffusion-prepared segmented gradient echo imaging, which also includes motion navigators and undersampling to improve efficiency and robustness for body imaging application (2,3) (Roccia et al, abstract accepted at ISMRM 2019). Here we propose to use diffusion at low b-values to assess perfusion based on Intra-Voxel Incoherent Motion (IVIM). IVIM has recently experienced increased interest, especially in the field of cancer imaging (4) where it has shown promising to characterize various tumour types (5) and assessing therapeutic effects (6,7). A key feature of IVIM perfusion imaging is that it does not involve contrast agents. However, there are technical challenges to overcome in particular for use in body imaging, such as artifacts from other bulk flow phenomena which we will address using our integrated motion navigators and optimised self-navigating acquisition trajectory. We further propose to extend the (low b-value) diff-prep acquisition sequence by interleaving it with an inversion recovery (IR)-prep module and to include a multi-echo kernel to achieve signal footprints that in addition to the low b-value perfusion, also encode T1 and T2* in a single scan. We will use advanced reconstruction methods incorporating MRI signal simulations and explore the use of convolutional neural networks (CNNs) to extract hypoxia vs perfused tissue information from the signal data.


1. O’Connor J, Boult J, Jamin Y, et al. Oxygen-enhanced MRI can accurately identify, quantify and map tumour hypoxia in preclinical models. Cancer Imaging 2015;15:P9. doi: 10.1186/1470-7330-15-S1-P9.
2. Roccia E, Vidya Shankar R, Neji R, Cruz G, Munoz C, Botnar R, Goh V, Prieto C, Dregely I. Accelerated 3D T 2 mapping with dictionary-based matching for prostate imaging. Magn. Reson. Med. 2019;81:1795–1805. doi: 10.1002/mrm.27540.
3. Shankar RV, Cruz G, Neji R, Roccia E, Botnar R, Goh V, Prieto C, Dregely I. Accelerated 3D 1 mm isotropic T2w-Imaging of the Prostate in less than 3 min. In: Proceedings 25th Scientific Meeting ISMRM, Paris, France. ; 2018.
4. Le Bihan D. Intravoxel incoherent motion perfusion MR imaging: a wake-up call. Radiology 2008;249:748–752. doi: 10.1148/radiol.2493081301.
5. Zhang Y-D, Wang Q, Wu C-J, Wang X-N, Zhang J, Liu H, Liu X-S, Shi H-B. The Histogram Analysis of Diffusion-Weighted Intravoxel Incoherent Motion (IVIM) Imaging for Differentiating the Gleason grade of Prostate Cancer. Eur. Radiol. 2015;25:994–1004. doi: 10.1007/s00330-014-3511-4.
6. Hauser T, Essig M, Jensen A, Laun FB, Münter M, Maier-Hein KH, Stieltjes B. Prediction of treatment response in head and neck carcinomas using IVIM-DWI: Evaluation of lymph node metastasis. Eur. J. Radiol. 2014;83:783–787. doi: 10.1016/J.EJRAD.2014.02.013.
7. Nougaret S, Vargas HA, Lakhman Y, et al. Intravoxel Incoherent Motion–derived Histogram Metrics for Assessment of Response after Combined Chemotherapy and Radiation Therapy in Rectal Cancer: Initial Experience and Comparison between Single-Section and Volumetric Analyses. Radiology 2016;280:446–454. doi: 10.1148/radiol.2016150702.

Back to Projects