Back to Projects

AI-enabled Imaging

Deep-learning PET-MR longitudinal reconstruction for lower-dose antibody-imaging in the understanding and treatment of cancer

Project ID: 2020_007

Student: Maxwell Buckmire-Monro

1st Supervisor: Andrew Reader, King’s College London
2nd Supervisor: Andy King, King’s College London
Clinical Champions: Vicky Goh and Gary Cook, King’s College London

Aim of the PhD Project:

This project explores new synergistic multi-modality data with an emphasis on AI-enhanced PET-MR image reconstruction methods, exploiting AI to improve imaging capabilities for cancer treatment monitoring. Methods will improve quantification, enable lower dose scanning and explore analyses that exploit synergies of information-rich longitudinal datasets.

Project Description / Background:

PET imaging is a powerful molecular and functional imaging technique which is in widespread use for diagnosing and assessing various diseases such as cancer, and which also finds application in cardiology and brain imaging [1, 2]. Of particular note more recently, antibody-based PET (Immuno-PET) is proving effective for visualising and characterising tumours through probe molecules [3].  Immuno-PET can help not only in deciding which patients can benefit from which type of cancer therapy, but also provides capabilities for monitoring efficacy of a particular treatment. Immuno-PET has focused on the use of a radiometal, zirconium 89 (89Zr), as it has excellent chemical and physical characteristics which are ideal for imaging of antibodies. The half life of 89Zr is 3.3 days, an ideal match for the relatively slow pharmacokinetics of some antibodies (2-3 days), but potentially more challenging for imaging therapeutic antibodies, which have a half life often in the range of 14-25 days. As a result, 89Zr-based immuno-PET imaging has now been used for a wide array of cancer-related targets including studies of relevant cell types in tissue. There is demonstrated ability to study drug target expression as well as monitor response to novel treatments. These methods therefore have potential to deliver molecular techniques for personalised treatment- matching the right drug with the right patient.

More generally, antibody-labelling techniques can be used for labelling drugs in their research phase, exploring and characterising pharmacokinetics over long time frames

Labelling of antibodies is now considered routine and standardised in some centres. Use of these techniques is expected to expand to new targets, new applications and generation of probes in more clinical centres. This is likely to increase the potential of the methods for drug development and routine clinical application to support personalised treatments.

However, there are drawbacks with 89Zr-based antibody imaging with PET: while the physical half life is ideal (3.3 days) for imaging antibodies, this also results in large radiation doses to imaging subjects / patients. For example, PET with 18F, with a half life of only ~2 hours, results in a dose of 5-10 mSv. If comparable levels of activity are used for 89Zr, then very broadly it is clear that dose will readily be an order of magnitude higher. As a consequence, there is a need to lower the administered activities as much as can be justified by the image quality required. The PET image reconstruction methodology is therefore crucial. Furthermore, 89Zr is not a pure positron emitter, which adds an unwanted background component to the measured PET data.

This project therefore aspires to the following progress:

  1. Development of advanced longitudinal PET image reconstruction algorithms, which are able to draw benefit from each and every longitudinal multi-modality scan of the subject under study
  2. Utilisation and co-modelling not only of the multiple PET datasets but also the longitudinal multi-contrast / multi-parametric MR data, with a view to direct multi-parametric synergistic PET-MR reconstruction from the rich multi-modality datasets
  3. Exploitation of deep learning methods to arrive at new state of the art longitudinal, multi-modal, multi-parametric and multi-data synergistic image reconstruction methods

We have already made initial advances in these image reconstruction methodologies, but this project will further develop and unify these advances, and importantly also for the first time embed the power of deep learning into longitudinal and multi-parametric reconstruction. We anticipate that using deep-learned longitudinal image reconstruction for multi-modal and multi-parametric imaging will result in more robust antibody imaging, delivering enhanced image metrics for greater capabilities in cancer imaging and personalised treatment planning and monitoring.

Expected academic background of the candidate: mathematics, physics, and/or computational science.


  1. Muehllehner, G. and J.S. Karp, Positron emission tomography. Physics in Medicine and Biology, 2006. 51(13): p. R117-R137.
  2. Jones, T. and E.A. Rabiner, The development, past achievements, and future directions of brain PET. Journal of Cerebral Blood Flow and Metabolism, 2012. 32(7): p. 1426-1454.
  3. van de Watering, F.C., et al., Zirconium-89 labeled antibodies: a new tool for molecular imaging in cancer patients. Biomed Res Int, 2014. 2014: p. 203601.
  4. Ellis, S. and A.J. Reader, Penalized maximum likelihood simultaneous longitudinal PET image reconstruction with difference-image priors. Med Phys, 2018.
  5. Mehranian, A., et al., PET image reconstruction using multi-parametric anato-functional priors. Phys Med Biol, 2017. 62(15): p. 5975-6007.
  6. Hammernik, K., et al., Learning a variational network for reconstruction of accelerated MRI data. Magn Reson Med, 2018. 79(6): p. 3055-3071.
  7. Reader, A.J. and J. Verhaeghe, 4D image reconstruction for emission tomography. Phys Med Biol, 2014.
  8. Mehranian, A., et al., Multi-modal synergistic PET and MR reconstruction using mutually weighted quadratic priors. Magn Reson Med, 2018.
  9. Mehranian, A and Reader A. J., Model-Based Deep Learning PET Image Reconstruction Using Forward-Backward Splitting Expectation Maximisation Oral presentation at the IEEE Medical Imaging Conference 2019

Back to Projects