内容来源：https://research.google/blog/spotlight-on-innovation-google-sponsored-data-science-for-health-ideathon-across-africa/

内容总结：

谷歌在非洲发起医疗AI创新挑战赛 本土团队用开源模型破解健康难题

近日，一场由谷歌赞助的“健康数据科学创意马拉松”在非洲大陆落下帷幕。来自非洲多国的研究团队利用谷歌开源医疗AI模型，针对宫颈癌筛查、孕产妇健康支持等本土迫切医疗需求，开发出一系列创新解决方案。

本次赛事与SisonkeBiotik、Ro'ya等非洲数据科学社区联合举办，吸引超过30支团队提交方案。参赛者基于谷歌开放的MedGemma、MedSigLIP等医疗AI模型进行开发，最终六支队伍凭借其创新性、可行性和本土化潜力进入决赛。

获奖项目充分体现“本土问题本土解决”的创新思路：冠军团队“Dawa Health”开发出支持多语言的宫颈癌教育筛查工具，助产士可通过WhatsApp上传影像，系统能实时识别癌前病变；季军项目“Mkunga”打造斯瓦希里语孕产健康咨询平台，致力于成为低成本远程医疗助手；特别奖项目“HexAI”则针对网络薄弱地区，开发出可离线使用的皮肤病筛查移动应用。

谷歌高级研究员表示，这类赛事反映出非洲科技社区正积极探索利用AI应对健康、农业、气候等领域的区域性挑战。目前部分获奖项目已进入试点阶段，其中冠军团队计划明年为5万名患者提供宫颈癌筛查服务。

此次赛事是谷歌“AI赋能非洲”倡议的组成部分，该倡议涵盖健康、教育、粮食安全等多个关键领域。谷歌表示将继续通过开放技术资源、提供计算支持等方式，助力非洲科技创新生态发展。

中文翻译：

创新聚焦：谷歌赞助的非洲健康数据科学创意马拉松

2025年12月12日
谷歌研究院高级研究科学家 梅西·阿西杜 与 塞库·雷米

非洲的研究人员和开发者们正运用谷歌的人工智能模型，应对关键的医疗挑战。

生成式人工智能正在迅速改变医疗健康领域，为更好地应对现实世界的健康挑战开辟了新机遇。在整个非洲大陆，从宫颈癌筛查到孕产妇健康支持，各界对应对此类挑战抱有广泛兴趣。

基于此背景，我们与三个泛非数据科学和机器学习社区——SisonkeBiotik、Ro'ya 以及 DS-I Africa——合作，举办了一场覆盖全非洲的“健康数据科学创意马拉松”。该活动聚焦于利用谷歌开放的健康AI模型，解决现实中的健康难题。

我们从超过30份雄心勃勃且技术含量高的提交方案中，依据其大胆的创意、扎实的基础以及对非洲卫生系统产生有意义影响的潜力，评选出了六支决赛团队。这些团队获得了全球专家的指导，以及谷歌研究院和谷歌DeepMind提供的技术资源支持。

这反映出非洲大陆各界对于如何利用人工智能为本地健康、农业及气候等优先领域创造本土解决方案的广泛兴趣。这也是谷歌在非洲更广泛的人工智能倡议的一部分，涵盖健康、教育、粮食安全、基础设施和语言等多个领域。

健康协同创新的新模式

创意马拉松，与黑客松类似，是一个跨学科团队为关键挑战设计解决方案的平台，既重视创意的创造性，也重视其可执行性。

该创意马拉松在卢旺达基加利举行的“深度学习Indaba”大会期间启动，作为“2025年非洲健康数据科学研讨会”的一部分。Indaba大会是非洲机器学习与人工智能社区的盛会，致力于加强和颂扬非洲AI发展。创意马拉松建立在增强非洲AI与健康社区能力的共同愿景之上。参与者还在研讨会上有机会参加了由谷歌研究院科学家塞库·雷米博士和梅西·阿西杜博士主导的MedGemma实践教程。

我们鼓励各团队探索谷歌开放的健康AI模型，即：MedGemma、TxGemma和MedSigLIP，并将这些模型应用于从诊断到政策框架等各类医疗健康挑战。评选标准包括创新性、可行性、情境相关性，以及在非洲环境中对谷歌AI工具的创造性运用。

从构想到影响：两阶段旅程

创意马拉松分两个阶段展开：

• 构思阶段：团队介绍成员，定义健康问题，并概述使用谷歌模型的AI驱动解决方案。入选团队获得导师指导、谷歌云Vertex AI计算积分以及技术文档，以完善其构想。
• 原型与路演阶段：团队提交展示其解决方案的演示视频，随后由专家评委小组评审，并选出六支决赛团队进行最终的现场路演。

决赛路演：创意与目标交汇

六支决赛团队在全球直播的“健康数据科学创意马拉松现场路演”中展示了他们的解决方案，随后是与评委的问答环节。活动不仅强调技术卓越性，也强调愿景、协作以及在非洲医疗健康领域的本土化创新。

获奖团队与奖项

以下是获奖项目，它们展示了谷歌的开放模型如何作为开发者构建解决方案的基础，帮助解决非洲紧迫的医疗健康挑战：

• 第一名兼观众选择奖：Dawa Health – AI驱动的多语言宫颈癌教育与筛查工具
  基于MedSigLIP构建，并利用Gemini RAG增强功能。助产士首先通过WhatsApp上传阴道镜图像。随后，基于MedSigLIP的分类器实时识别指示癌前病变/癌症的异常情况，同时Gemini根据世界卫生组织和赞比亚的诊疗规范提供情境化的临床指导[1, 2]。
  团队成员：Tafadzwa Munzwa, Khanyisile Magagula, Kudzal Mwedzi, Tariro Munzwa。

• 第二名：Solver – CerviScreen AI
  一个基于MedGemma-27B-IT的FastAPI网络应用，用于自动化宫颈细胞学筛查，该模型在CRIC数据集上使用LoRA进行了微调。可输出带标注的图像和简洁的临床建议，以辅助细胞病理学家。
  团队成员：Bonaventure F.P. Dossou, Ariane Houetohossou, Aurel Tchokponhoue, Ghilith Gbaguidi。

• 第三名：Mkunga – 孕产妇AI呼叫中心
  适配MedGemma和Gemini，利用TTS/STT技术提供斯瓦希里语的孕产健康建议，并部署在Vertex AI上。旨在扩展为低成本远程健康助手。
  团队成员：Malik Lanlokun, Beryl Apondi, Janet Njiru, Joseph Wacira。

• 最佳概念验证奖（离线/低连接环境）：HexAI – DermaDetect
  优先离线的移动应用，使社区卫生工作者能够使用适配后的设备端MedSigLIP和基于云的MedGemma进行高级分析，从而对皮肤病进行分诊，并创建一个用于持续改进的“数据飞轮”。
  团队成员：Ebrima S Jallow, Omar Keita。

• 最具趣味解决方案奖：MamaLens Lab
  为社区卫生工作者设计的支持多语言的离线安卓助手，适配MedGemma、MedSigLIP和TxGemma，用于评估英语和约鲁巴语的妊娠风险。
  团队成员：Ilerioluwakiiye Abolade, Faith Nchifor, Toyibat Adele, Simeon Krah。

未来展望

创意马拉松的结束标志着持续开发和规模化推广的开始。数支决赛团队正在探索如何扩展和部署他们的解决方案，并得到了项目期间建立的人脉网络的支持。冠军团队Dawa Health目前正在试点一项基于MedSigLIP分类器的宫颈癌筛查早期体验计划，并计划明年扩展到5万名患者，有效地展示了可扩展的现实世界影响力。

要了解更多关于谷歌开放模型的信息并开始探索它们以用于您自己的用例，请访问我们的HAI-DEF网站。我们欢迎您就模型及您的用例提供反馈。

致谢

本次倡议得以实现，离不开组织团队及合作伙伴的远见、协作与资金支持：Comfort Adesina（创意马拉松主席，SisonkeBiotik）、Abraham Owodunni（技术协调员，SisonkeBiotik）、Taliya Weinstein（创意马拉松主持兼顾问，SisonkeBiotik）、Ashery Mbilinyi（创意马拉松评委）、Lukman Ismaila（创意马拉松评委），以及来自谷歌研究院和DeepMind的：Mercy Asiedu、Sekou L. Remy、Sunny Jansen、Fereshteh Mahvar、Tiffany Chen、Yun Liu、Katherine Heller、Joelle Barrel、Richa Tiwari、Sunny Virmani。

英文来源：

Spotlight on innovation: Google-sponsored Data Science for Health Ideathon across Africa
December 12, 2025
Mercy Asiedu and Sekou Remy, Senior Research Scientists, Google Research
African researchers and developers used Google’s AI models to address crucial medical challenges.
Generative AI is rapidly changing healthcare, opening up opportunities to better address real-world health challenges. Across the African continent, there is broad interest in tackling such challenges, ranging from cervical cancer screening to maternal health support.
With the above in mind, in collaboration with three pan-African data science and machine learning communities — SisonkeBiotik, Ro’ya, and DS-I Africa — we hosted an Africa-wide Data Science for Health Ideathon, focused on leveraging Google’s open Health AI models to tackle real-world health challenges.
From over 30 ambitious and technically rich submissions, six finalist teams were selected for their bold ideas, strong foundations, and potential for meaningful impact across African health systems. These teams received mentorship from global experts and technical resources provided by Google Research and Google DeepMind.
This reflects broad interest across the African continent on understanding how AI can be used to create local solutions to local priorities around health, agriculture and climate. This is part of Google’s broader initiatives on AI for Africa including in areas of health, education, food security, infrastructure and languages.
A new model for collaborative health innovation
An ideathon, much like a hackathon, is a platform for interdisciplinary teams to design solutions to crucial challenges, valuing the idea creativity as much as its execution.
The Ideathon was launched during the 2025 Data Science for Health in Africa Workshop at the Deep Learning Indaba, the gathering of Africa’s machine learning (ML) and AI community, dedicated to strengthening and celebrating African AI, which took place in Kigali, Rwanda. The Ideathon built on the shared vision of building capacity across African AI and health communities. Participants also had the opportunity at the workshop to attend a hands-on MedGemma tutorial led by Dr. Sekou Remy and Dr. Mercy Asiedu, research scientists at Google Research.
Teams were encouraged to explore Google’s open health AI models, namely: MedGemma, TxGemma, and MedSigLIP, and apply these models to healthcare challenges ranging from diagnostics to policy frameworks. Submissions were judged on innovation, feasibility, contextual relevance, and the creative use of Google’s AI tools in African settings.
From idea to impact: A two-phase journey
The Ideathon unfolded in two phases:

• Idea development: Teams introduced their members, defined healthcare problems, and outlined AI-driven approaches using Google’s models. Selected teams received mentorship, Google Cloud Vertex AI compute credits, and technical documentation to refine their ideas.
• Prototype & pitch: Teams submitted demo videos showcasing their solutions, which were then reviewed by a panel of expert judges who chose six finalists for the final, live pitch.
  The final pitch: Creativity meets purpose
  Six finalist teams presented their solutions during the DS4H Ideathon Live Pitch, broadcast globally, followed by a team Q&A session with the judges. The event emphasized not only technical excellence but also vision, collaboration, and contextual innovation in African healthcare.
  Winning teams and awards
  Here are the winning projects that showcase how Google's open models can serve as a foundation on which developers may build and help solve pressing healthcare challenges in Africa:
  First place and Audience Choice Award: Dawa Health – AI-powered multilingual cervical-cancer education & screening tool
  Built on MedSigLIP and enhanced with Gemini RAG. Midwives first upload colposcopy images via WhatsApp. The MedSigLIP-based classifier then identifies abnormalities indicating precancer/cancer in real time, while Gemini provides contextual clinical guidance using WHO and Zambian protocols [1, 2].
  Team: Tafadzwa Munzwa, Khanyisile Magagula, Kudzal Mwedzi, Tariro Munzwa.
  Second place: Solver – CerviScreen AI
  A FastAPI web app for automated cervical-cytology screening based on MedGemma-27B-IT, fine-tuned with LoRA on the CRIC dataset. Outputs annotated images and concise clinical recommendations to assist cytopathologists.
  Team: Bonaventure F.P. Dossou, Ariane Houetohossou, Aurel Tchokponhoue, Ghilith Gbaguidi.
  Third place: Mkunga – Maternal AI call center
  Adapting MedGemma and Gemini to provide maternal health advice in Swahili using TTS/STT, deployed on Vertex AI. Aims to scale as a low-cost telehealth assistant.
  Team: Malik Lanlokun, Beryl Apondi, Janet Njiru, Joseph Wacira.
  Best proof-of-concept (offline / low-connectivity): HexAI – DermaDetect
  Offline-first mobile app that enables community health workers to triage skin conditions using adapted versions of on-device MedSigLIP and cloud-based MedGemma for advanced analysis, creating a “data flywheel” for continuous improvement.
  Team: Ebrima S Jallow, Omar Keita.
  Most fun solution: MamaLens Lab
  Multilingual offline Android assistant for community health workers, adapting MedGemma, MedSigLIP, and TxGemma to assess pregnancy risk in English and Yoruba.
  Team: Ilerioluwakiiye Abolade, Faith Nchifor, Toyibat Adele, Simeon Krah.
  What comes next
  The end of the Ideathon marks the beginning of sustained development and scaling. Several finalist teams are exploring ways to scale and deploy their solutions, supported by networks built during the program. The winning team, Dawa Health, is currently piloting an early access program for cervical cancer screening using a MedSigLIP-based classifier with plans to scale to 50,000 patients next year, effectively demonstrating scalable, real world impact.
  To learn more about Google's open models and start exploring them for your own use cases, visit our HAI-DEF site. We welcome your feedback on the models and your use cases.
  Acknowledgements
  This initiative was made possible through the vision, collaboration, and funding support of the organizing team and partners: Comfort Adesina (Ideathon Chair, SisonkeBiotik), Abraham Owodunni (Technical Coordinator, SisonkeBiotik), Taliya Weinstein (Ideathon Host and Advisor, SisonkeBiotik), Ashery Mbilinyi (Ideathon Judge), Lukman Ismaila (Ideathon Judge), and Google Research and DeepMind: Mercy Asiedu, Sekou L. Remy, Sunny Jansen, Fereshteh Mahvar, Tiffany Chen, Yun Liu, Katherine Heller, Joelle Barrel, Richa Tiwari, Sunny Virmani.