Early Detection Innovation Sandpit Award: Research Proposal Outline

Research Fellow

Dr Shuang Song

External investigators

Dr Patrick Murray (UCL), Dr Spencer Thomas (NPL)

In clinical, glioblastoma (GBM) is the most common primary adult brain tumour and carries one of the worst prognoses amongst human cancers, with a median survival time of about 15 months after diagnosis even with the best available treatment. Thus, the project aims at the early detection and prediction of glioblastoma recurrence through imaging.

By predicting the site and size of GBM recurrence and the trajectory of GBM recurrence, the developed model could be used to guide surgical and radiotherapy planning in future patients, in order to prolong the onset of recurrence as far as possible. In the project, images including T1-weighted imaging (pre/post gadolinium), T2-weighted images and diffusion imaging will be utilized.

Additionally, there are 3-5 time points for each patient, with each time point consisting of the MRI sequences outlined previously. Affine and non-linear registration techniques will be initially used for the co-registration of the MR images. Then, hybridisation of cycle-GANs and autoencoders for deep learning with multimodal data will be used to realize the spatio-temporal early detection.


Future prediction of tumour recurrence through computer modelling could have several uses including,

  1. targeted surgery and/or radiotherapy to the site at risk of recurrence;
  2. radiotherapy ‘dose painting’, whereby computer modelling can predict the locations at highest risk of recurrence allowing higher doses of radiotherapy to those sites whilst avoiding radiotherapy to the normal brain;
  3. it may be applicable to other cancers by allowing early detection of secondary tumours in radiotherapy fields.

A key strength of this project is that the model will use MR datasets which are routinely acquired clinically, and as such, after completion of this pilot project, roll-out to multiple centres throughout the country for a larger study is feasible.