AI system spots brain emergencies at record speed

Artificial intelligence capable of interpreting brain MRI scans within seconds is reshaping how hospitals triage neurological emergencies, following the development of a new model by researchers at the University of Michigan. The system analyses complex brain images almost instantly, flags cases requiring urgent intervention and identifies a broad spectrum of neurological conditions with accuracy levels reported to reach 97.5%, according to researchers involved in the project.

The technology addresses a persistent bottleneck in acute neurological care. Magnetic resonance imaging is a cornerstone of diagnosing stroke, tumours, haemorrhage and inflammatory brain diseases, yet scans often wait hours or longer for specialist review, particularly outside major medical centres. By contrast, the Michigan-developed model processes scans in seconds, ranking cases by urgency and highlighting abnormalities for radiologists and clinicians to assess.

Researchers say the system was trained on hundreds of thousands of real-world brain MRI scans paired with patient histories, enabling it to learn subtle patterns across age groups, disease types and imaging protocols. Unlike earlier tools designed to detect single conditions, such as large-vessel stroke or intracranial bleeding, this model evaluates scans holistically, mirroring how experienced neuroradiologists assess multiple possible diagnoses at once.

Clinical testing showed the system outperforming several advanced AI tools already in use or development, particularly in its ability to distinguish cases requiring immediate action from those suitable for routine review. The model demonstrated strong performance across a wide range of neurological findings, including tumours, demyelinating disease, hydrocephalus and vascular abnormalities, researchers said.

For emergency departments, speed is the central advantage. Stroke care, for example, operates on narrow therapeutic windows where minutes can determine outcomes. Automated prioritisation of MRI scans could allow clinicians to intervene faster, especially during peak hospital hours or overnight shifts when specialist coverage is thinner. In smaller hospitals without round-the-clock neuroradiology services, the technology could act as an early warning system, prompting rapid transfers or consultations.

Experts caution, however, that such tools are intended to augment, not replace, clinical judgement. The Michigan team emphasises that final diagnoses remain the responsibility of qualified physicians, with the AI serving as a decision-support layer that reduces delays and cognitive load. Radiologists reviewing early deployments reported that automated triage helped them focus attention where it was most needed, without disrupting existing workflows.

The development reflects a broader shift in medical imaging towards large-scale, multi-condition AI models trained on diverse datasets rather than narrow, task-specific algorithms. Previous generations of imaging AI often struggled when exposed to scans from different hospitals or scanners. By incorporating data from varied clinical settings and linking images to patient records, the new system aims to generalise more reliably across healthcare environments.

Regulatory and ethical considerations remain central as such technologies move closer to routine use. Issues around data privacy, bias and transparency are under scrutiny, particularly where AI influences urgent clinical decisions. The Michigan researchers say their training datasets were carefully curated and de-identified, and that ongoing evaluation is planned to monitor performance across different populations.

Healthcare administrators are also assessing cost and integration challenges. While AI triage promises efficiency gains, hospitals must invest in secure infrastructure and staff training to deploy such systems safely. Interoperability with existing imaging platforms and electronic health records is critical, as is clear accountability when AI recommendations conflict with human assessment.