NHK shows glaucoma exam in 3 seconds

NHK News on May 23 showcased two Japanese hospital uses of artificial intelligence that are already moving from research into clinical workflow: a glaucoma screening system shown reducing one exam from about 15 minutes to roughly 3 seconds, and an emergency-care model used to estimate how likely a patient is to deteriorate or die. The segment framed both as support tools for doctors rather than substitutes for them. That distinction runs through the institutions behind the systems as well. Researchers and developers say the software is meant to speed triage and flag risk, while physicians retain responsibility for diagnosis and treatment. ### How can a glaucoma check drop from 15 minutes to 3 seconds? Tohoku University professor Toru Nakazawa has been developing AI systems for early glaucoma detection, according to the university’s medical media site published on April 28. The university says glaucoma is the leading cause of blindness in Japan and that the disease often advances without obvious symptoms, making earlier detection important. (life.med.tohoku.ac.jp) The Tohoku University description says the project is built around three design choices: showing numerical evidence alongside the AI output, combining multiple models to improve early-stage detection, and making the model light enough to run on devices such as smartphones. The university says the aim is to build a system that can detect abnormalities in daily-life settings and eventually work with lower-cost equipment. (life.med.tohoku.ac.jp) ScienceDaily and Tohoku University previously described the same broader screening effort as a specialist-level glaucoma screening system designed for very fast checks from fundus images. Those reports said the project was intended for settings such as large-scale screenings and medically underserved areas. ### What is the emergency-room AI actually estimating? (life.med.tohoku.ac.jp) MeDiCU, an Osaka startup founded in 2023, developed an AI system introduced at Kansai Medical University Medical Center that predicts mortality probability, according to a Feb. 28 YTV News report carried by NTV. The report said the system learned from 160,000 cases and uses ICU patient vital signs and blood-test data to estimate the risk of sudden deterioration. (sciencedaily.com) YTV reported that the tool does not decide treatment policy. It is positioned as a support system that gives doctors one more input when making judgments, the report said. That matches the caution highlighted in the NHK segment about over-reliance on automated output. For context, a separate May 22 Forbes article about another emergency-medicine AI study said headlines claiming AI “beat” doctors overstated what the research showed. (news.ntv.co.jp) The article argued that performance claims in emergency care need to be read alongside workflow limits and clinical responsibility. (news.ntv.co.jp) ### Why are these examples getting attention now? May 23 coverage from NHK focused on applied uses rather than general claims about future AI in medicine. The glaucoma example showed a routine diagnostic task being compressed sharply in time, while the emergency-care example showed AI being used to rank risk and help prioritize attention. Tohoku University says its glaucoma work is also trying to address the “black box” problem by presenting evidence behind the output. (forbes.com) That is a practical governance issue in medicine, where speed alone is not enough for adoption. ### What are the limits the developers themselves are acknowledging? Tohoku University says its glaucoma system is intended to support early detection, not replace physician judgment. (life.med.tohoku.ac.jp) The university’s April 28 description says researchers are still working to improve the technology and expand use on cheaper devices and broader health platforms. YTV’s Feb. 28 report on the MeDiCU system was explicit that the emergency-care model is a support tool and not a machine that sets treatment. (life.med.tohoku.ac.jp) In both cases, the next step described by the developers is wider practical deployment with doctors still making the final call. (news.ntv.co.jp)