A research team from King's College London has developed an artificial intelligence model designed to help radiologists identify brain abnormalities in MRI scans. The study, published in Radiology AI, highlights the potential of this technology to address challenges such as radiologist shortages and increasing demand for MRI procedures.
The current backlog in radiology departments, caused by a lack of specialists and rising numbers of MRI requests over the past decade, can lead to delays in diagnosis and treatment. This is particularly significant for conditions like tumors, strokes, and aneurysms where timely intervention is crucial.
The new AI model was initially trained to distinguish between normal and abnormal brain scans. According to the researchers, it performed accurately when compared with expert radiologist assessments. Further testing involved presenting the model with previously unseen scans showing specific conditions including stroke, multiple sclerosis, and brain tumors; the model successfully recognized these abnormalities.
Unlike most existing AI models that rely on large datasets manually labeled by radiologists—a process that is both costly and time-consuming—the King's College London team created a system that learned from more than 60,000 existing brain MRI scans paired with their corresponding radiology reports. This approach eliminated the need for manual labeling by experts.
The researchers also designed the model so that it could respond to queries—such as "glioma," a type of brain tumor—by retrieving similar cases from its database. This function could assist with diagnostic reviews or be used as a teaching tool.
The study suggests that this AI system could be integrated into clinical workflows at the point of scanning. It could flag abnormal scans for further review, suggest findings to clinicians, detect possible errors in reports, or retrieve comparable cases from previous examinations. These capabilities have the potential to speed up diagnoses and reduce reporting delays.
"The next step is to run a randomised multicentre trial across the UK to see how abnormality detection improves workflows in practice. We are pleased to say that this trial will start in hospitals in 2026," commented Booth.
