Singapore’s brain-scan AI is helping urgent reports reach doctors sooner

At Singapore General Hospital, the most tangible early result from an AI brain-scan tool is a change in the reporting queue. The hospital says mean reporting time for positive intracranial haemorrhage cases fell from 34 to 27 minutes, while the share taking more than an hour dropped from 14.6% to 5%. The figures were reported in a 27 October interview with Healthcare IT News.[1]

The tool, iScreen, flags suspected bleeding on head CT scans and alerts clinicians so urgent images can be reviewed ahead of routine work. SGH developed the system with national health-technology agency Synapxe and Iota MedTech. Its September deployment announcement explicitly retained radiologists’ responsibility for checking findings and issuing final reports.[2]

The follow-up describes a two-month validation covering 3,475 scans. It reports 93.1% sensitivity and 98% specificity, with results and SMS alerts generated in an average of 3.8 minutes. An alert and a completed clinical report are different milestones; the faster alert figure must not be substituted for the report-turnaround measure.[1]

The benefit comes from reaching the right person

SGH’s official account describes a workload of approximately 50 head CT scans daily, around one in ten showing acute intracranial haemorrhage. The intended intervention is therefore not to remove clinical review but to help direct it towards the cases that may need it most urgently.[2]

That distinction explains why a workflow measure deserves attention alongside a model score. A system can recognise a concerning image yet have little practical effect if the result arrives in a separate interface nobody checks. Here, the alert is connected to the worklist and the clinicians responsible for responding.[2]

The reduction in reports taking over an hour is particularly useful to examine. An average alone can improve while a small group still waits a long time. Reporting both measures gives a clearer picture of timeliness, although it does not show which individual patients benefited or what happened after the report was issued.

A platform helps deliver the model, not its final verdict

iScreen is delivered through AimSG, the national medical-imaging AI platform. Synapxe describes that platform as enabling institutions to deploy models without tying themselves to one hardware or software supplier. The shared infrastructure is separate from the clinical judgement needed for each application.[3]

For another hospital, the transferable lesson is the link between detection, notification and action. Purchasing a model would not itself reproduce the same reporting improvement. Local staffing, the existing queue and the way alerts are acknowledged would all belong in the deployment evaluation.

The evidence here is hospital-reported operational experience, not a randomized trial demonstrating that iScreen caused better survival. Nonetheless, faster verified reporting is a concrete service improvement worth documenting. It gives clinicians earlier information while preserving a clear boundary: the AI prioritises attention; people remain responsible for the clinical conclusion.

Report an error or suggest a correction