Our impact on brain cancer research
Glioblastoma is a type of cancer that grows in the brain and spinal cord. Glioblastoma can occur at any age, but it is more common in those above the age of 50.
Support from our incredible donors has enabled research into the potential of using information from MRIs of patients with brain tumours (gliomas) to predict characteristics, including molecular phenotypes and future growth with artificial intelligence (AI).
Deep learning artificial intelligence (AI)
Project highlights
The team at Monash University’s world-class Department of Neuroscience, led by Professor Meng Law, Dr Jarrel Seah, Dr Andrew Dixon and Dr Jennifer Tang, have to date gathered, extracted, de-identified and catalogued 7.5 million images from 7,930 MRI studies from 690 patients.
This extensive collection of brain tumour MRI scans and corresponding pathological data is accessed to learn about the growth patterns and progression of glial series brain tumours in their natural history as well as following treatment.
“Distinguishing between progression and pseudo progression of brain tumours following immunotherapy, is more of an art than a science, and is an area of active research,” says Dr Jarrel Seah. “This research aims to build a foundational model capable of predicting temporal changes on MRI – a challenging goal which has been made possible by recent advances in latent diffusion models, which would help distinguish between progression and pseudo progression.”
The researchers intend to facilitate the use of unsupervised deep learning AI methods to enhance the detection and categorisation of gliomas and other types of brain tumours and are working on using the breakthroughs in this project to generate fully 3D MRIs.
Proof of concept models have been developed to generate 2D MRI slices as an intermediate step towards this goal. However, since the project’s initial conception, generative AI techniques have revolutionised, particularly with the development of latent diffusion models e.g. the models that power MidJourney and Stable Diffusion.
This project is also the basis for the infrastructure behind an Australia wide Brain Tumour Registry, which will include MRI scans from sites across Australia.
Imaging Informatics Platform (XNAT)
A dedicated research imaging informatics platform (XNAT) has been procured and installed, along with the necessary hardware for storage and computation. The team continues to work on applying latent diffusion models to product complete 3D MRI images.
In addition to the current project, the XNAT – with its ability to extract and de-identify large clinical imaging datasets – has been invaluable to many other projects seeking to apply deep learning in the space of medical imaging. For instance, it has been used to validate deep learning models used to identify common pathologies such as brain tumours on non-contrast CT brain, as well as train neural networks capable of segmenting perivascular spaces on MRI brain.
