AI in Medicine News — 2026-10-10 (Sat)

Overview

- Focus on diagnosis/imaging and safety/regulations. The expansion of autonomous skin cancer AI in the NHS and the government's acceptance of the UK's 44-item medical AI regulatory proposal. - Several studies have been reported warning against overreliance on image AI explanations (heat maps), and a real-world evaluation of AI for patients (BIDMC) has also been published. - In drug discovery, there are moves to introduce AI models, such as the partnership between GSK and Chai Discovery (after wet lab verification).

Diagnosis/Image

- In digital pathology, Aiforia and PathAI will partner to expand the provision of digital pathology AI tools. Measurement target and accuracy are not listed in the heading. - Regarding ultrasound and lung cancer, it was reported that an AI that uses regular ultrasound to detect lung cancer is comparable to a specialist radiologist (Bioengineer.org). Numerical values, target scale, and verification stage are not listed in the heading. - In tumor differentiation, an AI tool was reported to distinguish Ewing's sarcoma from similar tumors (Medical Xpress). Performance values are not listed. - As a warning against overconfidence, there is a study that warns against blindly trusting medical imaging AI (BusinessLine). The IIIT-H chest X-ray study has the same purpose (see Section 8). - Regarding chatbots for GPs, there is research that shows the use of AI chatbots to diagnose patients before consultation (The Independent). Verification stage is not described.

Clinical decision support

- Suki "Ask" was announced as a complex clinical decision support feature based on Wolters Kluwer UpToDate's expert AI. - Skin Analytics is reported to be the largest rollout of autonomous medical AI in NHS history. Targets skin cancer checks. - In head and neck cancer, AI was reported to integrate images and radiation doses to predict serious complications (Bioengineer.org). Performance values and verification stages are not listed. - There is a model that predicts cardiac risk from electrocardiograms from sleep tests (Mirage News). Performance values are not listed. - In Emergency Imaging, JFR discussed how to incorporate AI.

Generation AI and medical documents/medical records

- Healthleap is software for hospitals that uses AI to check hospital charts and detect undiagnosed inpatient risks. - EndoSoft Argus CareFlow is an AI for patient communication and follow-up calls that was announced at the ACG Annual Scientific Meeting. - In home medical care, an article was published that featured AI tools to meet the demand for home medical care (Modern Healthcare). - All are just headings, with no description of effectiveness or accuracy.

Hospital management/medical provision system

- NHS England South East expands Skin Analytics skin cancer AI screening to 18 trusts. - Regarding Medicare home health care, an article was published stating that AI discharge summaries cannot replace face-to-face consultations, which are a requirement for benefits. - Regarding the introduction of AI in the NHS, a document has been published that discusses ways to translate recommendations into implementation and enable safe and effective introduction. - There is an analysis that says that the valuation of medical AI is changing from obtaining approval to continuing revenue (Korea IT Times). - Newsweek's "Bad AI Recommendations Harm Sales" is not related to the medical care delivery system from the headline, and is therefore excluded from evaluation.

Drug discovery/clinical trials

- GSK and Chai Discovery implemented Chai's AI model after wet lab testing. The stage is considered to be exploratory/pre-clinical, but it is not specified in the heading. - Lonza and Onava are collaborating in the field of AI biopharmaceuticals with the aim of accelerating drug discovery. No stages are listed. - Nature "TechBio 3.0" discussed trends in closing the drug discovery loop with multimodal AI and generative chemistry. - In active learning, a method was reported that learns the cleavage site of molecules and guides drug discovery reactions (Bioengineer.org). It appears to be at the basic research stage. - There are no headings regarding clinical trial stages (e.g. Phase I).

Medical device approval/regulation

- In terms of guidelines and recommendations, the MHRA's AI in health recommendations were accepted by the UK government (UKAuthority). In a separate report, a 44-point blueprint for regulating medical AI was published. The relationship between the two cannot be determined from the headline. - In the United States, there are reports that executives from AI and tech companies are lobbying Trump administration health officials in government-run chat rooms (Modern Healthcare). - There are no headings regarding approval/designation of item units. Tempus AI stock articles are not included as they are related to market prices.

Safety/Error/Bias

- The IIIT-H study warned against blindly trusting medical AI heat maps. Targets X-rays. - BIDMC is the first to examine the safety and quality of patient-facing AI in primary care in a real-world setting. The content of the results is not listed in the heading. - Google AMIE reportedly could enhance the relationship between patients and doctors. Effect size and research design are not stated. - The UK's 44-point regulation proposal is a regulatory framework for medical AI.

Research results (major journals)

- The Lancet reported on a study showing promise for AMIE. Details are not listed in the heading. - JAMA has implemented measures against "AI slop" posts. - Nature investigated code sharing in clinical predictive modeling research in an AI-assisted scoping review. - The content of "Which AI Healthcare Features Are Ready to Use Today?" is unclear from the headline.

Japanese movements

- Kanazawa Medical University Hospital has begun full use of Ubee-generated AI in its electronic medical record environment. - Sakana AI has introduced the Japanese version of Sakana Namazu into its search tool for doctors, "Evidence Finder." - In a survey by the Ministry of Health, Labor and Welfare (White Paper 2026), over 70% of respondents responded positively to the use of medical AI. - A doctor-assistance chatbot was reported to be effective in 75% of cases in a trial run (Vietnam.vn). It is unclear from the headline whether this is a case in Japan or not, and the definition of "valid". - Kaunet's purchasing diagnosis AI will not be used because its relationship to medical care is unclear.

Remarks from VIPs

No relevant information (the list of important people being tracked is empty, and there are no individual comments in the headlines).