AI-powered portable MRI abnormality detection (APPMAD)
Lead Research Organisation:
King's College London
Department Name: Imaging & Biomedical Engineering
Abstract
We combine a range of research capabilities (new portable MRI scanner and our AI tools), and draw on multi- and interdisciplinary teams (clinicians and scientists from different faculties and different hospitals as well as a patient with relevant lived experience), to conduct early-stage translational medical research with the potential for patient benefit.
Two technological developments underpin the proposed work.
First, we developed an Artificial Intelligence (AI) tool that can accurately sort magnetic resonance imaging (MRI) brain scans into normal and abnormal (i.e., there appears to be disease). The AI "triage" tool that we built allows radiologists to report abnormal brain scans preferentially before normal scans which results in faster management of patients with disease. The downstream effect of our AI tool is to reduce the effects of disease, and related healthcare costs.
Second, recent technology allows MRI scans to be performed using a small, portable MRI scanner that does not require a dedicated hospital room with a fixed MRI scanner. Furthermore, unlike fixed MRI scans, it is also safe to use the portable MRI scanner next to metalwork. As such the portable MRI scanner can be used in GP surgeries, Community Diagnostic Hubs or wheeled to the bedside of a patient in an Intensive Care Unit who may be very unwell and therefore at high risk for transfer to the standard MRI department. The portable MRI scanner is also very cheap to buy and to run when compared to a fixed MRI. The "trade off" is that the images obtained are not as clear as the scans obtained in fixed MRI scanners. Nonetheless, the clarity of the images for relatively simple tasks such as sorting patients into normal and abnormal is sufficient - if abnormal, patients can be prioritised for onward referral for standard (fixed) MRI where the superior images can be used for more complex assessments.
Our aim is to use an AI trick called "transfer learning" to combine knowledge from our AI tool for triage (which was built for standard MRI scans) with a small number of research scans from the portable scanner in order to build an accurate portable MRI AI "triage" tool.
Our proposal will plausibly provide the initial evidence required to support the next translational research step that would bring the portable MRI AI "triage" tool to the clinic. Potential use-cases for translation are considerable. Because putting an intensive care patient inside a standard MRI scanner is both hazardous and laborious, a bedside portable scanner with an AI "triage" tool might indicate to the treating team whether it is sensible and necessary to proceed to standard MRI.
Additionally, the portable scanner would allow an initial triage in the community where patients have nonspecific clinical features. For example, many types of headache are a common problem but rarely associated with an abnormality. Community triage with an AI "triage" tool would plausibly allow more rapid onward referral for targeted imaging in those with an abnormality, for example, specialised MR imaging for possible brain tumour patients.
The research has immense potential to contribute to hospital and community medicine in countries like the UK. We also emphasise that low-income countries with almost no access to standard MRIs might benefit disproportionately from such a tool, as the portable scanner is cheap (~£200k compared to ~£1-2M for a standard MRI scanner).
Two technological developments underpin the proposed work.
First, we developed an Artificial Intelligence (AI) tool that can accurately sort magnetic resonance imaging (MRI) brain scans into normal and abnormal (i.e., there appears to be disease). The AI "triage" tool that we built allows radiologists to report abnormal brain scans preferentially before normal scans which results in faster management of patients with disease. The downstream effect of our AI tool is to reduce the effects of disease, and related healthcare costs.
Second, recent technology allows MRI scans to be performed using a small, portable MRI scanner that does not require a dedicated hospital room with a fixed MRI scanner. Furthermore, unlike fixed MRI scans, it is also safe to use the portable MRI scanner next to metalwork. As such the portable MRI scanner can be used in GP surgeries, Community Diagnostic Hubs or wheeled to the bedside of a patient in an Intensive Care Unit who may be very unwell and therefore at high risk for transfer to the standard MRI department. The portable MRI scanner is also very cheap to buy and to run when compared to a fixed MRI. The "trade off" is that the images obtained are not as clear as the scans obtained in fixed MRI scanners. Nonetheless, the clarity of the images for relatively simple tasks such as sorting patients into normal and abnormal is sufficient - if abnormal, patients can be prioritised for onward referral for standard (fixed) MRI where the superior images can be used for more complex assessments.
Our aim is to use an AI trick called "transfer learning" to combine knowledge from our AI tool for triage (which was built for standard MRI scans) with a small number of research scans from the portable scanner in order to build an accurate portable MRI AI "triage" tool.
Our proposal will plausibly provide the initial evidence required to support the next translational research step that would bring the portable MRI AI "triage" tool to the clinic. Potential use-cases for translation are considerable. Because putting an intensive care patient inside a standard MRI scanner is both hazardous and laborious, a bedside portable scanner with an AI "triage" tool might indicate to the treating team whether it is sensible and necessary to proceed to standard MRI.
Additionally, the portable scanner would allow an initial triage in the community where patients have nonspecific clinical features. For example, many types of headache are a common problem but rarely associated with an abnormality. Community triage with an AI "triage" tool would plausibly allow more rapid onward referral for targeted imaging in those with an abnormality, for example, specialised MR imaging for possible brain tumour patients.
The research has immense potential to contribute to hospital and community medicine in countries like the UK. We also emphasise that low-income countries with almost no access to standard MRIs might benefit disproportionately from such a tool, as the portable scanner is cheap (~£200k compared to ~£1-2M for a standard MRI scanner).
