{"id":78815,"date":"2026-09-26T16:20:22","date_gmt":"2026-09-26T20:20:22","guid":{"rendered":"https:\/\/www.globalvillagespace.com\/tech\/?p=78815"},"modified":"2026-09-26T16:20:22","modified_gmt":"2026-09-26T20:20:22","slug":"alibaba-radar-open-source-abdominal-ai","status":"publish","type":"post","link":"https:\/\/www.globalvillagespace.com\/tech\/alibaba-radar-open-source-abdominal-ai\/","title":{"rendered":"Alibaba open-sources RADAR: a generalist AI for 146 abdominal findings including cancers"},"content":{"rendered":"<p>Alibaba Group\u2019s research arm DAMO Academy has published a generalist medical imaging model called RADAR that its authors say can detect 146 abdominal clinical findings \u2014 including malignant tumours \u2014 from contrast\u2011enhanced computed tomography (CT) scans. The team released code and pre\u2011trained weights and published results in the journal Science, reporting strong diagnostic metrics and a radiologist comparison.<\/p>\n<h2>What DAMO RADAR is and what the team reports<\/h2>\n<p>The RADAR model is described by DAMO Academy as a vision\u2011language system trained on abdominal CTs and linked radiology reports. According to the research materials and the lab\u2019s public repository, RADAR was trained on a large dataset of contrast\u2011enhanced abdominal CTs paired with image\u2011text data. Multiple outlets and the DAMO repository report the training set size as over 400,000 scans; one summary cites the figure 424,911.<\/p>\n<p>Key performance figures the authors report:<\/p>\n<ul>\n<li>RADAR identifies 146 clinical findings spanning 18 abdominal anatomical structures.<\/li>\n<li>The model achieved a mean area under the curve (AUC) of 0.913 across 146 findings on roughly 40,000 real\u2011world examinations used for evaluation.<\/li>\n<li>On a subset of more than 27,000 emergency CT scans the AUC was reported as 0.904 by one summary of the paper.<\/li>\n<li>In a reader study involving 26 radiologists, RADAR\u2019s average performance exceeded that of 23 participants; with RADAR assisting readers, missed diagnoses were reportedly reduced by 10% and reading time fell by more than 30%.<\/li>\n<\/ul>\n<h2>What was released and licensing caveats<\/h2>\n<p>The DAMO team published code and model assets. Sources that inspected the official repository state the distribution is split by license:<\/p>\n<ul>\n<li>Code for training and inference: Apache\u20112.0 (per the repository README), which permits commercial use.<\/li>\n<li>Pre\u2011trained weights and auxiliary data: CC BY\u2011NC\u2011SA 4.0 (attribution, non\u2011commercial, share\u2011alike) per the public distribution on Hugging Face.<\/li>\n<\/ul>\n<p>One independent commentator noted this means the procedural recipe for building RADAR is open, while the ready\u2011to\u2011use pre\u2011trained weights are restricted to non\u2011commercial use under the CC BY\u2011NC\u2011SA 4.0 terms. That distinction matters for hospitals, startups and cloud providers planning to deploy or resell inference services.<\/p>\n<h2>How RADAR\u2019s approach differs from many previous models<\/h2>\n<p>RADAR\u2019s authors link images to free\u2011text clinical reports rather than relying on manual, per\u2011image annotation. According to the repository description and reporting, the vision\u2011language training aligns CT volumes with report text and converts CT data into three\u2011dimensional anatomical units for more precise report alignment. The team positions this as a scalable method to cover many findings without labor\u2011intensive pixel\u2011level labelling.<\/p>\n<h2>Context and immediate implications<\/h2>\n<p>If reproduced independently, a generalist model that reads hundreds of thousands of CTs and flags diverse findings could reduce the need for dozens of disease\u2011specific models. The DAMO team suggests the method could extend to other imaging modalities. The published AUCs\u20140.913 mean across 146 findings\u2014are substantially higher than the single\u2011task baseline figure one commenter cited (0.776) on the same benchmark, but that is an internal comparison and requires external validation.<\/p>\n<h2>What remains unresolved and what to watch for next<\/h2>\n<ul>\n<li>Regulatory clearance and clinical integration: published model performance in a paper and repository does not equate to medical device approval in jurisdictions such as the United States; the announcement and repository materials do not document regulatory submissions or approvals.<\/li>\n<li>External reproduction: the author community flagged that independent remeasurements on data from other countries are needed to validate generalization beyond the reported test sets. One commentator suggested independent reproductions would surface within weeks of release.<\/li>\n<li>Commercial terms: while code is Apache\u20112.0, the CC BY\u2011NC\u2011SA 4.0 license on weights restricts direct commercial use\u2014an important operational constraint for companies and cloud providers seeking to offer RADAR\u2011based services.<\/li>\n<\/ul>\n<h2>Practical checklist for hospitals or developers evaluating RADAR today<\/h2>\n<ol>\n<li>Confirm which assets you need: training code vs pre\u2011trained weights. Apache\u20112.0 code can be used commercially; Hugging Face weights carry CC BY\u2011NC\u2011SA 4.0 restrictions.<\/li>\n<li>Run local reproduction tests: obtain the evaluation scripts from the repository and measure RADAR on in\u2011house, de\u2011identified CT datasets before any clinical trial or deployment.<\/li>\n<li>Assess regulatory pathway early: consult local medical device and health data regulators about evidence required to use RADAR for clinical decision support in your country.<\/li>\n<li>Plan workflow integration: test how RADAR outputs integrate with PACS and reporting systems and whether the model\u2019s findings map to your radiology report taxonomy.<\/li>\n<li>Document performance drift checks: establish periodic re\u2011evaluation on new local cases and emergency CT subcohorts to track AUC and false negatives.<\/li>\n<\/ol>\n<h2>Value added beyond the press release<\/h2>\n<p>This article compares reported licensing\u2014Apache\u20112.0 for code versus CC BY\u2011NC\u2011SA 4.0 for weights\u2014and highlights practical implications: open procedural reproducibility but restricted commercial reuse of the published pre\u2011trained model. It also gives a short operational checklist hospitals and developers can follow immediately to validate and plan regulatory and integration steps prior to any clinical use.<\/p>\n<h2>Sources and attribution<\/h2>\n<p>The technical and performance claims in this article are drawn from DAMO Academy\u2019s public repository and the RADAR paper reported in Science and summarized by multiple outlets. Specific reported figures\u2014&#8221;mean AUC 0.913 across 146 findings,&#8221; &#8220;about 40,000 real\u2011world examinations,&#8221; &#8220;trained on over 400,000\/424,911 scans,&#8221; &#8220;outperformed 23 of 26 radiologists,&#8221; and the licensing split (Apache\u20112.0 code; CC BY\u2011NC\u2011SA 4.0 weights)\u2014are attributed to the project\u2019s public materials and independent commentary cited in reporting.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>DAMO Academy published RADAR\u2014a vision\u2011language model trained on contrast\u2011enhanced CT scans that the team says detects 146 abdominal findings with an average AUC of 0.913 and whose code and weights have been released under mixed licenses.<\/p>\n","protected":false},"author":1,"featured_media":78817,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6023],"tags":[6031,6034,7035,7034,6036,7036],"class_list":["post-78815","post","type-post","status-publish","format-standard","has-post-thumbnail","category-latest","tag-ai","tag-alibaba","tag-damo-academy","tag-medical-imaging","tag-open-source","tag-radiology"],"_links":{"self":[{"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/posts\/78815","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/comments?post=78815"}],"version-history":[{"count":1,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/posts\/78815\/revisions"}],"predecessor-version":[{"id":78816,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/posts\/78815\/revisions\/78816"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/media\/78817"}],"wp:attachment":[{"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/media?parent=78815"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/categories?post=78815"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.globalvillagespace.com\/tech\/wp-json\/wp\/v2\/tags?post=78815"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}