A deep learning tool which can automatically identify abnormalities on a brain MRI scan has been backed by £1million for its further development.
The tool would enable the immediate automated triage of abnormalities matching that of a consultant neuroradiologist, and will help to address the fact that 330,000 patients currently wait more than 30 days for their MRI reports in the UK.
That number is forecast to increase further due to greater demand for MRI and the availability of radiologists – a situation in the UK which is mirrored globally.
Researchers from King’s College London say they have developed the means to detect more than 90 distinct abnormalities when testing their algorithm while selecting the scans that are normal.
Their work has now been given support from the Medical Research Council Development Pathway Funding Scheme (MRC DPFS), with the award of £1million to progress its technology.
The work follows earlier findings that it is possible to automate brain MRI image labelling where the researchers found that more than 100,000 exams can be labelled in less than 30 minutes.
This was an important step to overcome a bottleneck to label scans at scale. Scans labelled at scale allowed researchers to build a hugely promising brain MRI abnormality detection triage tool.
Without a large number of labelled scans, a deep learning tool could not have learnt the categories of normal or abnormal with such accuracy.
Dr Tom Booth, senior lecturer in neuroimaging at the School of Biomedical Engineering & Imaging Sciences and Consultant Diagnostic and Interventional Neuroradiologist at King’s College Hospital, and Professor Sebastien Ourselin, head of School of Biomedical Engineering & Imaging Sciences, lead the project, which has been building up since its first Health Research Authority/Research Ethics Committee approval in 2018.
Dr Booth said simulations show this triage tool would triage effectively for outpatient scans at two London hospitals and reduce the time to report abnormal scans.
“Specifically, abnormal scans can be reported in five days rather than nine days in one large hospital and 14 days rather than 28 days in another large hospital,” he said.
The researchers will now look to improve performance accuracy and ensure generalisability, where the algorithm works well with new data in new hospitals, across the UK.
To achieve generalisability, the researchers are ingesting data across the UK in order to finesse the model and ensure it works with high accuracy in different hospitals, with different scan manufacturers and imaging sequences.
The research is a strong first step towards automating the triage process.
“Immediate triage of a brain MRI into normal or abnormal allows early intervention to improve short and long-term clinical outcomes,” Dr Booth said.
“Detecting illnesses earlier in the patient pathway would result in lower costs for the healthcare system given that less specialised medicine, and fewer hours of treatment, are needed for patient recovery.”
