[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-damo-radar-open-ct-model":3,"topics-all":39,"news-related-c77dc26f-954d-4208-8341-c83099ccfd8f":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":25,"news_slug":32,"published_at":33,"created_at":34,"modified_at":35,"is_published":36,"publish_type":37,"image_url":15,"view_count":38},"c77dc26f-954d-4208-8341-c83099ccfd8f","达摩院开源医学影像模型:146种病症,胜过多数放射科医生","阿里达摩院联合浙大一院等机构研发的通用医疗影像模型 DAMO RADAR 登上 Science:读一张腹部增强 CT 即可筛查 146 种病症、覆盖 18 个器官;近 4 万例真实检查平均 AUC 0.913,读片研究胜过 26 位放射科医生中的 23 位,代码与权重已开放。","一张腹部增强 CT 躺进扫描仪,放射科医生要逐个器官排查;而 DAMO RADAR 一次读完全腹,给出 146 项病症的置信度评分——这是阿里达摩院联合浙江大学医学院附属第一医院等机构研发的通用医疗影像模型,9 月 18 日登上 Science,代码与权重同步开放([人民网健康](http:\u002F\u002Fhealth.people.com.cn\u002Fn1\u002F2026\u002F0918\u002Fc14739-40801368.html))。\n\n## 从单病种到「一次读全腹」\n\n医疗影像 AI 过去十年的主流做法是一个模型盯一种病:肺结节、乳腺癌筛查各有专用模型,部署成本高,长尾病症没人做。RADAR 走的是通用路线:输入增强腹部 CT 和器官分割掩码,输出一张 146 列的 CSV,每列对应一个「器官-病症」对的置信度,覆盖肝、胰腺、胃、结直肠等 18 个器官,包括这四类癌症([SCMP](https:\u002F\u002Fwww.scmp.com\u002Ftech\u002Fbig-tech\u002Farticle\u002F3368055\u002Falibaba-open-sources-medical-ai-model-can-detect-cancer-and-nearly-150-conditions))。阿里将其称为「世界首个专家级通用医学影像模型」——这是官方自述,独立同行还没盖章。\n\n## 1500 万份报告当标注\n\n训练数据是 40 万+ 次腹部增强 CT 检查和 1500 万个「图像-报告」对。关键在标注:团队没有雇医生逐层画框,而是让模型对齐影像和医生已写好的报告——报告本身就是标签。这条对齐路线把标注成本从人力密集压到数据整理,也让覆盖面从十几种扩到 146 种。\n\n## 数字与许可证都要看细\n\n近 4 万例真实检查上,146 项发现的平均 AUC 为 0.913;读片研究里,它平均表现超过 26 位参试放射科医生中的 23 位;作为辅助工具使用时,医生漏诊率下降约 10%,诊断时间缩短超过 30%([NDTV](https:\u002F\u002Fwww.ndtvprofit.com\u002Fscience\u002Falibaba-s-medical-ai-outperforms-radiologists-in-detecting-cancers-and-other-conditions-across-18-organs-12067608))。\n\n开源成色有个容易被忽略的细节:GitHub 仓库代码采用 Apache-2.0 许可,但模型权重单独以 CC BY-NC-SA 4.0 发布——研究可用,商用需要另行授权([TechTimes](https:\u002F\u002Fwww.techtimes.com\u002Farticles\u002F327749\u002F20260919\u002Falibaba-radiology-ai-outperforms-23-26-radiologists-across-146-diseases-science.htm))。第三方实测显示推理显存约 16GB,单张消费级 GPU 就能跑([MindStudio](https:\u002F\u002Fwww.mindstudio.ai\u002Fblog\u002Falibaba-radar-medical-ai-cancer-detection)),仓库地址:[alibaba-damo-academy\u002Fdamo-radar](https:\u002F\u002Fgithub.com\u002Falibaba-damo-academy\u002Fdamo-radar)。\n\n## 所以呢\n\n权重开放的直接受益者不是三甲医院,而是请不起按次付费 API 的县医院和基层诊所——一台 GPU 工作站就能常驻一个「第二读片人」。但 NC 许可证意味着商业化路径仍握在阿里手里,欧美医院落地还要过独立人群验证和监管审批。开源医学影像模型的竞赛,正从「能不能做」进入「谁用得起」。","https:\u002F\u002Fgithub.com\u002Falibaba-damo-academy\u002Fdamo-radar","12f049bc-e83f-4bb4-a8e2-239791c4dcf9",[11,16,19,22],{"id":12,"name":13,"slug":13,"description":14,"color":15},"9112951a-2abb-4214-b63a-385ec7afb2ba","ai-for-science","AI for Science 专题：追踪 AI 在生命科学、化学材料、物理世界模型等科学方向的关键突破",null,{"id":17,"name":18,"slug":18,"description":15,"color":15},"e676a5cf-1f24-472f-a765-86fa21a1bc3c","ai-model",{"id":20,"name":21,"slug":21,"description":15,"color":15},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":23,"name":24,"slug":24,"description":15,"color":15},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[26],{"id":27,"lang":28,"title":29,"summary":30,"content":31},"8ce37ad4-25b1-49e8-8ba1-afb4d656b7f8","en","DAMO RADAR: Alibaba's open CT model beats 23 of 26 radiologists","Alibaba DAMO's Science-published RADAR reads one abdominal CT for 146 findings across 18 organs; AUC 0.913, beat 23 of 26 radiologists. Code and weights open.","A single contrast-enhanced abdominal CT used to mean a radiologist working through organ after organ. DAMO RADAR reads the whole abdomen in one pass and returns confidence scores for 146 findings — a generalist medical imaging model developed by Alibaba's DAMO Academy with the First Affiliated Hospital of Zhejiang University's medical school and other institutions, published in Science on September 18 with code and weights released ([People's Health](http:\u002F\u002Fhealth.people.com.cn\u002Fn1\u002F2026\u002F0918\u002Fc14739-40801368.html)).\n\n## From single-disease tools to one full-abdomen read\n\nFor a decade, medical imaging AI mostly meant one model per disease: dedicated lung-nodule detectors, dedicated breast screening stacks — costly to deploy, with long-tail conditions left uncovered. RADAR takes the generalist route: it takes a contrast-enhanced abdominal CT plus an organ segmentation mask as input, and outputs a 146-column CSV where each column is a confidence score for one organ-condition pair, covering 18 organs including the liver, pancreas, stomach and colorectum, and the cancers that arise there ([SCMP](https:\u002F\u002Fwww.scmp.com\u002Ftech\u002Fbig-tech\u002Farticle\u002F3368055\u002Falibaba-open-sources-medical-ai-model-can-detect-cancer-and-nearly-150-conditions)). Alibaba calls it \"the world's first expert-level generalist medical imaging model\" — that is the company's own description, not an independently endorsed fact.\n\n## 15 million reports as labels\n\nTraining used more than 400,000 abdominal CT exams and 15 million image-report pairs. The interesting part is annotation: instead of hiring clinicians to delineate findings scan by scan, the team aligned imaging with radiology reports doctors had already written — the reports themselves act as labels. This vision-language alignment approach compresses annotation cost from labor-intensive to data-engineering, and it is why coverage reaches 146 conditions rather than a dozen.\n\n## Read the numbers and the license closely\n\nAcross nearly 40,000 real-world exams, RADAR averaged an AUC of 0.913 over the 146 findings; in a reader study it outperformed, on average, 23 of the 26 participating radiologists; used as an assistive tool, it cut missed diagnoses by about 10% and shortened diagnosis time by more than 30% ([NDTV Profit](https:\u002F\u002Fwww.ndtvprofit.com\u002Fscience\u002Falibaba-s-medical-ai-outperforms-radiologists-in-detecting-cancers-and-other-conditions-across-18-organs-12067608)).\n\nOne easily missed detail in the release: the GitHub repository code is Apache-2.0, but the model weights are separately published under CC BY-NC-SA 4.0 — research use allowed, commercial deployment requires a separate agreement ([TechTimes](https:\u002F\u002Fwww.techtimes.com\u002Farticles\u002F327749\u002F20260919\u002Falibaba-radiology-ai-outperforms-23-26-radiologists-across-146-diseases-science.htm)). Third-party hands-on testing reports roughly 16GB of VRAM during inference, within reach of a single consumer GPU ([MindStudio](https:\u002F\u002Fwww.mindstudio.ai\u002Fblog\u002Falibaba-radar-medical-ai-cancer-detection)); repository: [alibaba-damo-academy\u002Fdamo-radar](https:\u002F\u002Fgithub.com\u002Falibaba-damo-academy\u002Fdamo-radar).\n\n## So what\n\nThe immediate beneficiaries of open weights are not top-tier hospitals but county hospitals and clinics that cannot afford per-scan diagnostic APIs — one GPU workstation can host a permanent second reader. But the NC license keeps the commercial path in Alibaba's hands, and Western deployments still face independent demographic validation and regulatory review. The open medical-imaging race is shifting from \"can it be done\" to \"who can afford to run it\".","damo-radar-open-ct-model","2026-09-20T21:09:28Z","2026-09-20T21:09:33.096735Z","2026-09-20T21:09:33.096743Z",true,"agent",1,[40,48],{"slug":13,"tag_slug":13,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":36,"created_at":46,"modified_at":47},"AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":36,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"b9898ba4-65f5-4d80-b9ab-338b01fbd685","NASA 与 IBM 开源月球模型:极区找冰误差降 22%","nasa-ibm-lunar-foundation-model","2026-09-18T13:10:34+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"c99751d5-418e-49d5-99d3-e43b84c80ec7","IBM与NASA开源月球基础模型:Lunar Foundation Model","nasa-ibm-lunar-foundation-model-sombench","2026-09-19T09:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"dc2f4ead-963c-4a8e-bd41-400bebf83bb4","物理、几何、外观一个模型全包:Puffin-World 开源,相机 roll 误差低至 0.26°","puffin-world-native-3d-world-states","2026-09-06T19:09:41+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"7cc1b87c-fe06-495a-9c01-9516d0c16354","腾讯混元 HunyuanImage-3.0 全面开源：80B 总参 \u002F 13B 激活的自回归 MoE，把多模态理解和生图拉到同一框架","tencent-hunyuanimage-3-moe-autoregressive","2026-08-05T01:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"009ab609-27bf-4c34-92a6-2ce53eb25b69","Qwen 3.5 原生多模态新思路：DeepStack Vision Transformer 多层特征融合解析","qwen-3-5-deepstack-vit-multilayer","2026-05-09T13:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"3de0fc02-b400-4c9a-a613-426b5004b27b","NASA 和 IBM 的 Prithvi：首个在轨 AI 地理空间基础模型开启遥感新范式","nasa-ibm-prithvi-on-orbit-geospatial","2026-05-07T11:00:00+00:00"]