Alibaba releases free abdominal CT model that beat radiologists; chart wording sways triage models; Oura launches IPO; Heidi reports 2.8 million visits a week
Alibaba's Hangzhou-based research arm, DAMO Academy, published a generalist abdominal CT model in Science on Thursday and released the code and weights on GitHub for anyone to download. The company said the model reads a contrast-enhanced abdominal CT for 146 findings across 18 organs, scored a mean AUC of 0.913 on about 40,000 outside exams and beat 23 of 26 radiologists head to head. The figures are the company's own, reviewed by Science; the model was validated on Chinese patients only and has no clearance from the Food and Drug Administration (FDA) or any other regulator. Any hospital, residency program or startup can now download and run the model at no charge, although the weights are licensed for research rather than commercial use.
Two papers in npj Digital Medicine addressed language and authorship in clinical AI. Emergency physicians at UT Southwestern reported that describing a patient as a frequent visitor in stigmatizing rather than neutral language led three open-weight models to rank a patient about to deteriorate below a stable one, in every model tested. A St. Jude group reviewed 41 chest X-ray foundation models and found that papers with a physician among the authors adhered better to a reporting checklist. Separately, Oura launched its initial public offering on Monday morning, 50 million shares at $40 to $44 on the Nasdaq, with a prospectus that describes a wellness product rather than a medical device and does not mention the physicians to whom the company began routing members two weeks ago.
- Evidence Sept 18, 2026The model was trained on more than 400,000 contrast-enhanced abdominal CT exams paired with their reports, about 15 million anatomy-tagged image-text pairs, according to the GitHub page, which links to the Science paper. The South China Morning Post reported Thursday that the model was tested on nearly 40,000 real-world exams and recorded a mean AUC of 0.913 across 146 findings in 18 organs, cancers included. The test set came from eight outside centers, the model outperformed 23 of 26 radiologists, and radiologists using it as an assistant gained about 10 points of sensitivity and read more than 30% faster, TechTimes reported; the code is released under the Apache 2.0 license and the weights are licensed for research, not commercial use. The Science paper is behind a paywall, and the figures here come from the company's page and the press reports. The model was validated on Chinese patients only and on contrast-enhanced abdominal CT only, with no prospective trial and no regulatory clearance in any country. Under a final order published Thursday in the Federal Register, every radiology detection, diagnosis and triage algorithm in the U.S. still requires its own 510(k) clearance from the FDA, open weights or not.
- Evidence Sept 19, 2026Emergency physicians and a bioinformatician at UT Southwestern built pairs of patients from two academic emergency departments, matched on Emergency Severity Index, in which one patient deteriorated within six hours and the other did not, then asked Gemma, Qwen and DeepSeek which to see first, before and after inserting stigmatizing or neutral versions of 18 attributes. Stigmatizing language about frequent emergency department use pushed the deteriorating patient down the list in every model and dataset, by as much as 9.4%, and neutral phrasing scored 1.4 to 7.1 points higher, according to the paper in npj Digital Medicine. Psychiatric history moved rankings by up to 9.5% but reached significance in only one model, and race, language and insurance showed no consistent harm. The authors concluded that the wording of the input is a safety-critical design choice. Thursday's post covered a model that scored identical presentations as less severe when the patient was female.Source: npj Digital Medicine
- Money Sept 21, 2026The offering consists of 13.5 million shares from the company and 36.5 million from existing holders, listed on the Nasdaq under the symbol OURA, with Goldman Sachs, Morgan Stanley and J.P. Morgan leading; at the midpoint of the range the deal would total about $2.1 billion, less than the $3 billion at a valuation above $16 billion that Bloomberg reported in August. The S-1 shows 5.0 million paid members at June 30, revenue of $1.21 billion in the nine months to June 30 (hardware $974 million, membership $240 million, up 74%) and year-to-date net income of $60.8 million; Fierce Healthcare reported the member demographics, about 72% women and 27% over 45. The filing describes an AI health companion, Oura Advisor, and custom health LLMs, and states that the company markets its products as general wellness products that it believes are not subject to FDA device rules. The filing does not mention Counsel Health, ACCESS or Medicare, although Oura members have been able to reach Counsel's physicians inside the app since Sept. 17.
- Evidence Sept 21, 2026Radiologists at St. Jude reviewed every chest radiograph foundation model published through March 2025, 41 in all, and scored each paper against CLAIM, the reporting checklist for medical imaging AI. The brief communication in npj Digital Medicine found transformer architectures and language-model components becoming dominant, model sharing still uncommon and one variable that tracked with quality: publications with a physician among the authors adhered to CLAIM significantly better and discussed fairness more often. The paper is short, and its authors report the link as a correlation only.Source: npj Digital Medicine
- Deployment Sept 20, 2026Darran Foo, Asia-Pacific chief medical information officer at Heidi, told the HIMSS Asia-Pacific meeting in Singapore that the company's ambient scribe supports 2.8 million patient visits a week across 190 countries and 110 languages, and that a New Zealand emergency department cut documentation time from 17 minutes to four per patient. The figures are the vendor's, presented from a conference stage and unaudited, and no published study was cited for the New Zealand result. Foo also cited a survey Heidi published in July of 1,823 clinicians in 25 countries: 86% said they use AI daily or several times a week, 83% said they do so without employer guidance or a formal policy, and 68% named hallucination as their top concern. The 83% figure is consistent with an Imprivata survey of U.S. health system leaders released last week and covered in the Sept. 15 post, in which 72% reported AI deployed without IT approval.