Research Update

What are we funding?

In 2019, The Aftershock committed to support research within the Department of Neuroscience at Monash University (CCS, Alfred Hospital) led by Professor Meng Law, Dr Jarrel Seah, Dr Andrew Dixon and Dr Jennifer Tang. The research aims to show the potential of using information from MRIs of patients with brain tumours (gliomas) for the prediction of characteristics and future growth with artificial intelligence. The current project aims to collect a large dataset of brain tumours on MRI with associated pathology results, and apply unsupervised deep learning techniques to improve the diagnosis and classification of gliomas and other brain tumours.

Progress to date

To date the research team has focused on the collection and curation of available MRI brain datasets at the two institutions that the primary investigators are at, Alfred Health and The Royal Melbourne Hospital. A Victorian multi-site ethics application was submitted and approved, along with the site-specific applications and governance for the two sites, Alfred Health and The Royal Melbourne Hospital.

Unfortunately, a significant delay was encountered due to the COVID-19 pandemic in Melbourne last year, with many ethics committees deferring review of non COVID-19 research projects. Since then, all required permissions have been obtained and data transfer from both sites has commenced. Since the previous update, the additional dedicated hardware including an additional two Quadro RTX 8000 graphics processing units as well as another 100TB of hard disk space has been procured and installed to store and process the incoming data with local redundancy to protect against data loss. MRI data for the relevant patient cohort dating back to 2010 has been extracted from the Royal Melbourne Hospital, and similar extraction is under way at Alfred Health.

Challenges faced to date include the difficulty of obtaining appropriate ethical and governance approvals at each site during COVID-19, the logistical challenges of shifting large datasets as well as the heterogeneity of the data, as different sites have used different MRI scanning protocols at different time points.

Next steps

Following collation and patient reconciliation of both datasets from Alfred Health and the Royal Melbourne Hospital, the researchers will conduct preliminary experiments to check convergence of the candidate unsupervised models on the obtained data. Using the recently released RSNA-MICCAI brain tumour radiogenomic classification challenge data, this will form a useful test set to evaluate the central hypothesis of the project – that temporally based unsupervised learning is capable of providing additional information when attempting to predict clinical features relevant to the diagnosis and management of brain tumours.

The project that The Aftershock began supporting 2 years ago, now has joined 2 large GLOBAL AI challenges with RSNA-MICCAI-ASNR-Kaggle in partnership with the American Society of Neuroradiology and Medical Image Computer and Computer Assisted Interventions. These experiments will allow the researchers to determine the incremental benefit of additional unsupervised data, and determine future directions, such as whether further data extraction will be required from additional sites, or from earlier years.

Due to the challenges of COVID-19, the project timeline has been revised with the team now looking to start model development that will continue until December 2021. The availability of an external dataset from the RSNA-MICCAI challenge now means that we can perform our external validation in a retrospective rather than a prospective manner. This is expected to take about six months, so the final completion of the project is expected in mid to end 2022.