NHK: AI cuts glaucoma test to 3 seconds

NHK News said on May 23 that Japanese researchers and startups are testing AI tools for two different clinical bottlenecks: glaucoma exams that can take about 15 minutes, and emergency decisions that must be made before a patient reaches a hospital. The broadcaster’s report said one system cut a glaucoma-related test to about three seconds in trials and another was being used to estimate death risk in emergency settings. NHK’s own story could not be directly accessed because of site restrictions, but the underlying research and pilot programs it appears to reference are documented by the companies and institutions involved. The glaucoma item lines up with work by DeepEyeVision, a Jichi Medical University spinout that said on September 11, 2024 it had begun developing AI to estimate visual-field values from OCT eye scans. The company said standard visual-field testing used in glaucoma care takes about 15 minutes per session and places a burden on both patients and medical staff. DeepEyeVision said its goal is to use deep learning on OCT images to reduce testing time substantially and support treatment planning. (deepeyevision.com) ### How does the glaucoma system appear to work? DeepEyeVision said its program uses OCT, or optical coherence tomography, to capture cross-sectional images of the eye and then applies AI to infer values normally obtained through a visual-field exam. In its September 2024 release, the company said the project is aimed at improving early glaucoma diagnosis and easing the burden of repeated testing. It also said it had started technical exchanges with Jichi Medical University, the University of Tsukuba and OCT manufacturers, and had brought in ophthalmologist Makoto Furuyama as an adviser. (deepeyevision.com) A separate March 2025 release from Tohoku University shows a parallel track in Japan: a specialist-level glaucoma screening AI based on fundus images, published in *npj Digital Medicine*. Tohoku said glaucoma is the leading cause of irreversible blindness in Japan and worldwide and described its system as a screening tool that could operate within seconds. That is not the same project NHK referred to, but it shows the broader push in Japan to move glaucoma detection from long, specialist-led workflows toward faster AI screening. (deepeyevision.com) ### What is the emergency-care AI actually doing? Science Tokyo said on December 12, 2025 that its researchers had developed a prehospital AI triage system for head-trauma patients and would start a demonstration project on December 15 with Kaga City, Kaga Medical Center and the Kaga City Fire Department. The university said the system uses information that emergency responders can gather at the scene — including age, consciousness level, blood pressure and pupil status — to predict intracranial bleeding and help choose the most appropriate destination hospital before arrival. (tohoku.ac.jp) The Science Tokyo team said the underlying machine-learning model, based on XGBoost, predicted traumatic intracranial hemorrhage with sensitivity and specificity of about 75%. The university said researchers later added a function to predict more severe cases and built a web application that can run on tablets and smartphones. Kaga City said the system is intended to help responders decide from the scene whether a patient should be taken directly to a hospital capable of advanced treatment. (isct.ac.jp) ### Did NHK overstate the “death risk” angle? NHK’s wording about “death risk” could not be independently matched in the accessible source documents, so that part should be treated cautiously. The closest verified material is a Japan Science and Technology Agency review document and Science Tokyo’s release, both of which frame the emergency AI around preventing treatment delays in severe head trauma and reducing avoidable deaths through faster triage and transport decisions. (isct.ac.jp) In one case example included in the JST material, a 70-year-old traffic-accident patient deteriorated after transfer and later died, illustrating the problem the system is designed to address. ### What comes next? December 15, 2025 is the clearest dated milestone in the verified record: that is when the Kaga City emergency pilot began, according to Science Tokyo and the city. On the eye-care side, DeepEyeVision said its next step is social implementation of the OCT-based glaucoma program with academic and industry partners, but it did not give a commercialization date in the release reviewed here. (isct.ac.jp) (shingi.jst.go.jp)