机器之心发布 机器之心编辑部 2025 年 9 月 17 日，中国科学院香港创新研究院人工智能与机器人创新中心（ CAIR ）在香港正式开源发布其最新科研成果—— EchoCare "聆音"超声 基座大模型（简称"聆音"）。该模型基于超过 450 万张、 涵盖 50 多个人体器官的大规模 超声影像数据集训练而成，在器官识别、器官分割、病灶分类 等 10 余项典型超声医学 任务测试中表现卓越，性能全面登顶。同时，"聆音"已在山东大学齐鲁医院、中南大学湘雅医院、香港中文大学医学院的多个超 声检查领域完成 3000 多例临床回溯性验证，与当前 SOTA （最优）模型相比，性能平均提高 3%~5% 。 "聆音"首创的结构化对比自监督学习框架，高效解决了传统超声 AI 模型普遍存在的技术难题，包括对大规模标注数据的依赖、难以处理长尾分布问题、模 型泛化能力不足以及缺乏足够领域知识。同时，该框架为 AI 大模型在其他医疗领域的应用探索提供了创新路径，是人工智能与临床医学深度融合的又一里 程碑式突破。 超声 AI 的发展困局：数据枷锁与泛化死结 超声诊断凭借动态成像与无创安全的双重优势，已成为医学影像领域的"核心工具"。目 ...

Core Insights - The fourth Global Digital Trade Expo has commenced in Hangzhou, highlighting the importance of digital trade and cooperation between countries [1] Group 1: Indonesia's Role - Indonesia is recognized as a significant driver of digital economy growth in Southeast Asia, particularly in e-commerce and fintech sectors [1] - The Indonesian government expresses a strong desire to collaborate with China in market expansion, technology development, and talent cultivation [1] Group 2: UAE's Participation - The UAE aims to enhance its leading position in digital trade and services through participation in the expo, viewing it as a strategic opportunity for collaboration with China and other nations [1] - The UAE ambassador to China emphasizes the potential for new prospects in cooperation and experience exchange between the two countries [1] Group 3: Exhibitor Highlights - The Indonesian national pavilion features 15 companies from sectors such as telecommunications, logistics, fintech, e-commerce, and economic special zones, showcasing their digital achievements [1] - The UAE national pavilion includes cutting-edge areas like artificial intelligence, digital entertainment, cybersecurity, cloud services, smart logistics, and medical AI, presenting a multifaceted view of its digital ecosystem [1]

Market News - US stock indices collectively declined, with the Dow Jones down 0.37%, Nasdaq down 0.33%, and S&P 500 down 0.28%. Notable tech stocks had mixed performances, with Intel rising over 6% due to news of seeking investment from Apple, while Tesla increased nearly 4%. Chinese concept stocks saw gains, with the Nasdaq Golden Dragon China Index up 2.83% [1] - International precious metal futures generally fell, with COMEX gold futures down 1.24% at $3768.5 per ounce and COMEX silver futures down 1.11% at $44.115 per ounce. In contrast, international oil prices surged, with WTI crude oil up 2.21% and Brent crude oil up 2.02% at $68.32 per barrel [1] Industry Insights - Alibaba announced a collaboration with Nvidia for Physical AI, focusing on data synthesis, model training, and environment simulation. Alibaba's CEO stated that achieving AGI is a certainty, with a long-term goal of developing ASI. The company is investing 380 billion yuan in AI infrastructure over three years, expecting a tenfold increase in energy consumption by 2032 compared to 2022 [2] - The Chinese government is supporting the establishment of international data service businesses in specific regions, which is expected to increase demand for AI storage. Cloud service providers are projected to raise capital expenditures by $18 billion in FY2025, boosting AI storage needs [3] - The Beidou satellite navigation and positioning service industry is projected to reach a total output value of 575.8 billion yuan by 2024, with expectations of over 400 million new Beidou-supported devices by the end of 2028 [3] - The commercial space industry in China is expected to see explosive growth, with market size projected to increase from 9.2 billion yuan in 2020 to 310 billion yuan by 2024, reflecting a compound annual growth rate exceeding 100% [4] - The first Medical Artificial Intelligence Conference (MAIC2025) will focus on AI applications in healthcare, highlighting the transformative impact of AI technologies in the medical field [5] - The Chinese government is actively promoting AI in healthcare through various initiatives, including the establishment of innovation funds and the development of high-quality data sets for AI applications [6]

Group 1 - The first Medical Artificial Intelligence Conference (MAIC2025) will open on September 26 in Shandong, focusing on the integration of AI in the healthcare sector and aligning with China's health and innovation strategies [1] - The conference is positioned as a key platform for capital and industry connection amid the rapid expansion of the medical AI market [1] - The event's location in Jinan highlights the regional industrial layout, as Jinan is a core area for the transformation of new and old growth drivers, building an "AI + healthcare" industry cluster [1] Group 2 - The National Health Commission's guidelines on the classification of AI medical software products and the establishment of innovation funds across various regions are supporting the development of medical AI [1] - Wanda Information will showcase its latest achievements and typical application cases in the field of medical artificial intelligence, focusing on "AI + health management" [1] - Weining Health will present a series of innovative products, including the doctor-exclusive AI workstation WiNBOT, the next-generation smart hospital system WiNEX, the medical large model WiNGPT, and the healthcare intelligent assistant WiNEX Copilot [1]

Core Viewpoint - Yidu Technology (医渡科技) has won a bid for a Phase III clinical research project for the recombinant human nerve growth factor (SMR001) eye drops, with a total project amount of approximately RMB 55,817,597.00 [1] Group 1: Company Overview - Yidu Technology is positioned as a driver of AI medical transformation in China, with a mission to make precision medicine accessible to everyone [1] - The company develops professional, efficient, precise, and inclusive medical AI products and solutions based on its proprietary core algorithm engine, YiduCore [1] Group 2: AI Medical Value Proposition - The core of AI in healthcare is to enhance the quality and efficiency of medical evidence, optimizing research, diagnosis, and treatment processes [1] - Yidu Technology aims to improve the efficiency and accessibility of medical services, providing doctors with precise decision-making tools, shortening the distance from molecules to patients for pharmaceutical companies, and offering affordable precision diagnosis and treatment for patients [1] - The value of medical AI is reflected in the respect for individual lives and the safeguarding of health and well-being [1]

Core Viewpoint - Geoffrey Hinton, a Turing Award and Nobel Prize winner, has shifted his perspective on AI, now viewing it as a symbiotic relationship rather than a threat, particularly in the medical imaging field [1] Group 1: AI in Medical Imaging - AI is transforming disease screening, diagnosis, risk assessment, and clinical decision-making in the medical imaging market in China [1] - The company United Imaging established a subsidiary, United Imaging Intelligence, in 2017, focusing on AI medical solutions, and has launched over 100 AI applications, with numerous certifications from NMPA, FDA, and CE [1] - AI-assisted diagnosis is now a common tool for radiologists, significantly reducing the rate of missed diagnoses [3] Group 2: Key Personnel and Contributions - Gao Yaozong, the Chief Scientist at United Imaging Intelligence, has a background in computer vision and AI, previously working at Apple before returning to China to focus on medical AI [2][18] - Gao emphasizes the greater value of AI in healthcare compared to entertainment, highlighting the urgent need for AI solutions in China's medical landscape [2] Group 3: AI Development and Applications - The company has developed a lung nodule diagnostic grading system, C-Lung-RADS, based on extensive data from Chinese populations, enhancing early lung cancer screening accuracy [4] - United Imaging has created a mobile health management unit that provides lung cancer screenings to underserved areas, successfully identifying early-stage lung cancer cases [4] - The company has also launched an intelligent electronic medical record system that significantly reduces the time doctors spend on documentation [4][17] Group 4: Future Directions and Challenges - The ideal future technology path combines the strengths of general large models and specialized small models to enhance disease recognition and ensure precision in critical tasks [4][15] - The company faces challenges in developing truly universal, cross-modal medical imaging models and effectively integrating multi-modal information [12][13] - Regulatory challenges exist as AI medical products are classified as high-risk and require stringent approval processes [13][14] Group 5: Collaboration and Data Utilization - The company collaborates with hospitals to gather data while ensuring patient privacy and data security, employing a "data does not leave the hospital" approach [9] - Partnerships with leading hospitals are prioritized to ensure high-quality data for model training, with plans for multi-center validation for broader application [10] Group 6: Market Reach and Deployment - United Imaging's AI applications have been deployed in over 4,000 hospitals globally, integrating AI into imaging devices and providing independent AI platforms for various medical scenarios [11]

Core Viewpoint - Geoffrey Hinton, a prominent figure in AI, has shifted from warning about AI risks to expressing optimism about its applications, particularly in medical imaging, where AI can outperform human doctors in information retrieval and risk assessment [1] Company Overview - United Imaging Healthcare established a subsidiary, United Imaging Intelligence, in 2017 to focus on AI medical solutions, leading to the launch of over 100 AI applications, with 15 approved by NMPA, 15 by FDA, and 31 by EU CE, making it a leader in global medical AI certifications [1] - The company has developed a comprehensive ecosystem combining imaging devices and AI technology, which is attractive for the medical AI market in China [3][19] Key Personnel - Gao Yaozong, the Chief Scientist and Senior Vice President of United Imaging Intelligence, has a background in computer vision and AI, previously working at Apple before returning to China to contribute to the medical AI sector [2][19] Market Dynamics - The Chinese medical imaging market is undergoing transformation due to AI, which is enhancing disease screening, diagnosis, risk assessment, and clinical decision-making [1] - The vast population and diverse disease spectrum in China provide a rich data environment for training AI models, making it an ideal location for medical AI development [19] AI Applications in Healthcare - AI-assisted diagnosis is becoming a common tool for radiologists, significantly reducing the rate of missed diagnoses by serving as a "second pair of eyes" [3] - United Imaging has developed a lung nodule diagnosis grading system, C-Lung-RADS, based on 120,000 cases of Chinese population data, improving early lung cancer screening accuracy [4] Technological Innovations - The company employs a dual-path strategy of using both open-source models and proprietary development to enhance AI capabilities in medical imaging [6] - During the COVID-19 pandemic, the company rapidly developed AI systems for diagnostic support, demonstrating strong technical responsiveness [8] Future Directions - The ideal future technology path involves combining the strengths of general large models and specialized small models to enhance disease recognition and ensure precision in critical tasks [15] - The company aims to make AI a supportive tool for doctors, automating routine tasks and providing diagnostic suggestions, while addressing ethical and responsibility issues for higher autonomy in AI [16] Collaboration and Data Management - United Imaging collaborates with hospitals to gather data while ensuring patient privacy and data security, employing a "data does not leave the hospital" approach for model training [9] - The company focuses on multi-center validation to ensure the generalizability of AI models across different hospitals [10] Regulatory Environment - AI medical products are classified as high-risk and require stringent regulatory approval, with over 100 AI products already approved in China [14] - The company actively participates in shaping regulatory guidelines and industry standards to facilitate the development of AI in healthcare [14]

Core Insights - AI-assisted diagnosis has become a common tool for radiologists, significantly reducing the rate of missed diagnoses by serving as a "second pair of eyes" for doctors [1] Group 1: AI in Medical Imaging - AI technology is increasingly integrated into daily practices of radiologists, enhancing diagnostic accuracy [1] - The development of AI applications based on local data is crucial for adapting to Chinese clinical settings [1] Group 2: C-Lung-RADS System - A new lung nodule diagnostic grading system, C-Lung-RADS, has been proposed by Union Medical and West China Hospital, utilizing 120,000 cases of lung imaging data from the Chinese population [1] - This system integrates multi-dimensional information, including imaging data, clinical records, and follow-up changes in nodules, to provide a comprehensive risk assessment for lung nodules [1] - The implementation of C-Lung-RADS has improved the precision and clinical applicability of early lung cancer screening [1]

Core Insights - Shanghai Yinghe Medical Technology Co., Ltd. (Yinghe Yimai), incubated by Yimai Sunshine, achieved first place in the CT image automatic report generation task and third place in the 18-label abnormal classification task at the MICCAI 2025 3D Visual Language Multimodal Challenge, showcasing China's leading capabilities in the field of medical imaging AI [1][3][7] Group 1: Competition Overview - The MICCAI 2025 challenge featured participation from top institutions such as Harvard, Stanford, and Johns Hopkins, highlighting its international and authoritative nature [2] - The challenge is viewed as a "touchstone" for measuring the practical application capabilities of AI technology in clinical diagnostics [2] Group 2: Technical Achievements - Yinghe Yimai ranked first among 161 participants in the CT image automatic report generation task, achieving an average BLEU score of 0.2791, which is 5 points higher than the baseline [3] - In the multi-label abnormal classification task, Yinghe Yimai ranked in the top three among 118 participants, with an average AUC of 0.846 for 18 types of chest abnormalities [3] Group 3: Industry Integration - Yimai Sunshine's deep integration of "industry + AI" has accelerated the practical application of medical imaging AI, providing a vast amount of real-world imaging data for model training [4] - The collaboration creates a closed-loop ecosystem of "data-model-application-data," enhancing both parties' positions in the medical imaging AI sector [4] Group 4: Technological Advancements - Yinghe Yimai launched the world's first multimodal full-process medical imaging base model "Yinghe Mijia," which shortens the AI product development cycle by over 70% [5] - The company collaborates with major tech firms like Huawei and Alibaba to expand application scenarios beyond imaging diagnostics to health management and insurance services [6] Group 5: Future Directions - Yimai Sunshine is building the largest and fastest-growing medical imaging database in China, with nearly 10 million new cases added annually, enhancing the quality and standardization of data [4] - The recent dual breakthroughs in international competition and practical applications position Yinghe Yimai as a leader in China's medical imaging AI technology [7]

Core Viewpoint - The establishment of the National AI Application Pilot Base in the medical field at the 2025 China International Service Trade Fair marks a significant step in promoting AI applications in healthcare, aiming to enhance efficiency and accessibility in medical services [1][7]. Group 1: AI Application and Services - The pilot base serves as a large "incubation factory" where hospitals and enterprises can access comprehensive services covering diagnosis and drug development [1][7]. - During the fair, 31 AI medical interactive products were showcased, including a system for early warning of eye diseases and major chronic conditions, which attracted significant public interest [1][4]. - The AI eye disease and chronic disease warning system can automatically generate analysis reports within three minutes, identifying over 30 eye diseases and 10 systemic risks [4][5]. Group 2: Infrastructure and Development - The pilot base aims to create a unified support platform for the medical industry, facilitating the standardized and scalable application of AI technologies [7][9]. - It focuses on two main areas: precision diagnosis and biopharmaceutical manufacturing, providing a comprehensive service system that includes product development and application promotion [9][10]. - The base is supported by the Beijing Municipal Development and Reform Commission and the Beijing Municipal Health Commission, with operations managed by a dedicated company [7][9]. Group 3: Data Challenges and Solutions - Data management is identified as a significant challenge, with issues related to ownership, transaction mechanisms, and compliance requirements [10][11]. - The pilot base is working on creating a trusted data space to facilitate low-cost and efficient data usage for hospitals and AI companies [10][11]. - A medical big model and application evaluation system is being developed to provide standardized assessments of AI applications in real clinical settings [11][12]. Group 4: Future Vision - The ultimate goal of the pilot base is to leverage AI to improve public health outcomes by enabling early detection and intervention through data monitoring [12]. - The vision includes a seamless connection between medical institutions and enterprises, allowing for proactive health management and reducing the incidence of diseases [12].