DeepGEM大模型

Search documents
从“看图识癌”到“读片知基因” 金域医学、腾讯、广医附一院联合开发病理基因多模态大模型
Zheng Quan Ri Bao Wang· 2025-10-12 13:21
本报讯 (记者丁蓉)在人工智能(AI)的支持下,通过常规组织病理图像即可预测肿瘤患者基因突变 情况,正在成为现实。2025年10月11日,广州金域医学(603882)检验集团股份有限公司(以下简 称"金域医学")、腾讯控股有限公司(以下简称"腾讯")和广州医科大学附属第一医院国家呼吸医学中 心对外宣布,三方将依托AI大模型DeepGEM,共同开发病理基因多模态大模型,为更多患者提供准 确、及时且普惠的基因突变预测新技术。 据悉,DeepGEM大模型由广州医科大学附属第一医院及广州呼吸健康研究院、腾讯共同开发而成,相 关成果已发表于国际顶尖期刊《柳叶刀·肿瘤》,实现了从"看图识癌"到"读片知基因"的突破。金域医 学的加入,不仅为DeepGEM大模型用于肺癌的诊断提供了更多病例的验证以及临床服务场景,还将进 一步联合推动病理基因多模态大模型的开发,实现具有泛化性的多部位、多癌种、多组学的AI辅助诊 断。 肿瘤基因突变预测的重大突破 在规范的癌症诊疗流程中,病理诊断是确诊癌症的基础,基因测序则是精准诊疗的必要前提。然而,常 规的基因检测方法(二代高通量测序NGS)技术复杂、耗时长和成本高,难以广泛应用于临床,尤其是 ...
传统肿瘤基因检测价格过万,大模型能改变多少
Di Yi Cai Jing· 2025-10-12 03:17
为破解这一临床痛点,广州医科大学附属第一医院及广州呼吸健康研究院何建行/梁文华教授团队联合 腾讯生命科学实验室姚建华教授团队,开发出人工智能大模型"DeepGEM",实现了利用常规组织病理 图像来预测肺癌基因突变。该项目的多中心数据集验证结果显示,DeepGEM大模型能够提供准确、及 时且经济的基因突变及其空间分布的预测,在多种常见肺癌驱动基因突变的预测1分钟即可完成,精准 度达78%~99%。 为进一步验证DeepGEM大模型的临床价值,2025年,DeepGEM大模型研发团队联合第三方医检机构金 域医学,对该模型进行了更大规模的验证。 这次金域医学与腾讯、广州医科大学附属第一医院和广州呼吸健康研究院也正式签署了协议,就 DeepGEM模型的深度开发与应用达成合作。 传统的肿瘤基因价格昂贵,导致难以普及,而大模型出现后,试图改变这种困境。 10月11日下午,金域医学、腾讯和广州医科大学附属第一医院国家呼吸医学中心对外宣布,三方将依托 AI大模型DeepGEM,共同开发病理基因多模态大模型,为更多癌种患者提供准确、及时且普惠的基因 突变预测新技术。 被视为"癌中之王"的肺癌,在所有癌种中基因突变最多。基因检测 ...
金域医学:联合腾讯、广医附一院开发病理基因多模态大模型
Zheng Quan Shi Bao Wang· 2025-10-11 10:39
Core Insights - The collaboration between Kingmed Medical, Tencent, and Guangzhou Medical University First Affiliated Hospital aims to develop the AI model DeepGEM for predicting gene mutations in cancer patients using routine pathological images [1][5][6] - DeepGEM has demonstrated a high accuracy rate of 78% to 99% in predicting common lung cancer driver gene mutations, significantly improving the efficiency and accessibility of genetic diagnostics [2][4] Group 1: Development and Technology - DeepGEM is developed by a collaboration between Guangzhou Medical University First Affiliated Hospital, Guangzhou Respiratory Health Research Institute, and Tencent, marking a significant advancement from traditional pathology to genetic insights [1][3] - The model utilizes innovative techniques such as Multiple Instance Learning (MIL) and an end-to-end architecture that enhances prediction accuracy without the need for manual tumor region annotation [3][4] Group 2: Clinical Application and Validation - Kingmed Medical is providing a large-scale dataset for validating DeepGEM, with over 15,000 NGS tests conducted annually and a sample size of 4,260 lung cancer patients across various medical institutions [4][5] - The model has reached clinical auxiliary diagnostic levels for identifying mutations in genes like EGFR, KRAS, and ALK, showcasing its robustness and compatibility for clinical use [4][6] Group 3: Future Prospects and Expansion - The partnership aims to expand the application of DeepGEM beyond lung cancer to other cancer types, integrating various omics data for a comprehensive diagnostic approach [5][6] - The collaboration is seen as a milestone in the exploration of AI-driven pathology-genetics models, with aspirations to enhance the efficiency of clinical research and diagnostics in both cancer and rare diseases [6]