Magnetic resonance Imaging abnormality Deep learning Identification (MIDI)

Radiology workloads soared over the past decade due to changing demographics, increased screening and updated clinical guidelines specifying imaging in the clinical pathway. However, only 2% of NHS trusts met their 2019 reporting requirements. Before COVID, at any one time, 330,000 patients wait >30 days for their MRI reports - and there were forecasts that this would increase due to more demand than radiologist availability.
Following COVID there has been a 210% increase in the imaging backlog.

Reporting delays cause poorer short- and long-term clinical outcomes with late detection of illness inflating healthcare costs in accordance with World Health Organisation findings. This is exacerbated by radiologists routinely reporting from the back of report queues, introducing systemic delay in identifying abnormalities.

Our project makes a tool that will immediately flag brain MRI abnormalities allowing radiology departments to prioritise limited resources into reporting abnormal scans, thereby expediting early intervention from the referring clinical team, improving clinical outcomes and lowering healthcare costs. Furthermore, there will likely be reductions in mean-scan-reporting-time, through accurate and reliable clinical decision support, and reporting backlogs. Specific examples relate to cancer, stroke or infection where an early diagnosis can allow early treatment.

To achieve this we propose a deep learning decision-making tool for brain MRI scans that is both accurate and usable across the UK.

A clinically validated decision-making tool to identify abnormalities on brain MRI scans does not exist. This project meets this clinical need by developing such an abnormality detection tool using deep learning, enabling immediate automated triage of abnormalities with target product profile diagnostic performance matching that of a consultant neuroradiologist. This is important as 330,000 patients wait >30 days for their MRI reports in the UK. The number is forecast to increase, due to more demand for MRI than the availability of radiologists to report these scans. UK-specific workforce shortages in radiology are negatively impacting patient care by delaying diagnosis; a similar picture is seen globally.
Immediate triage of a brain MRI into normal/abnormal allows early intervention to improve short- & long-term clinical outcomes. Furthermore, detecting illnesses early would result in lower costs for the healthcare system given less specialised medicine & fewer hours of treatment are needed for patient recovery.
The proposal is enabled by another deep learning labelling tool, allowing MRI scans to be labelled at scale. Our publications confirm that the core approach is not only feasible & effective but also allows subcategorization into relevant groups as secondary objectives.
However, there is a technology gap that must be filled to provide a capability ready for a clinical trial.
A key objective is to complete analytical validation by 30 months in multiple prospective external test sets.
Preparation for Class 2a Medical Device regulatory approval of software & phase III trial design, including a health economic evaluation, will be achieved within 36 months. A trial, beyond the current Award, will complete clinical validation.
The commercial exploitation pathway will be developed during the Award prior to the trial where a 3rd party (i.e. commercial partner) will be responsible. IP relates to the software we will generate for the triage tool in the proposal.