In this article
Healthcare leaders face unprecedented challenges: rising labor costs, clinician workforce shortages, and the pressure to deliver safer, faster, and more personalized care. Microsoft’s AI technologies represent a leap forward in how organizations can automate clinical workflows, extract actionable insights from multimodal data, and drive performance. As part of this effort, Microsoft is introducing proprietary healthcare AI models, now available in Microsoft Foundry, and healthcare AI model evaluator, a user-friendly tool available for download on GitHub that allows teams to compare and assess AI model performance.
Fueling innovation with a growing healthcare AI model catalog
In spring 2024, Microsoft introduced healthcare AI models in Foundry—a collection of cutting-edge open source multimodal medical imaging models specifically designed for testing, fine-tuning, and building AI solutions tailored to healthcare-specific tasks. Along with partner offerings, Microsoft introduced three foundation models—MedImageInsight, MedImageParse, and CXRReportGen—with the goal of minimizing the compute and data requirements typically needed to build multimodal models from scratch. A year and a half later, the catalog now features more than 50 first-party and third-party models for a range of modalities, including radiology, pathology, dermatology, protein, and more.
Now, Microsoft is introducing two new proprietary AI models, MedImageInsight Premium and CXRReportGen Premium that set new benchmarks for accuracy and sensitivity in clinical imaging.
MedImageInsight Premium supports imaging applications across X-rays, MRIs, dermatology, and pathology, delivering up to 7–15% higher accuracy and requiring less fine-tuning than its open-source counterpart. It allows for image quality assessment, classification, search, metadata analysis, outlier detection, and supports regulatory workflows such as institutional review board submissions for FDA compliance. CXRReportGen Premium, optimized for chest X-ray report generation, has been further trained with massive amounts of real-world data to help create higher quality reports more similar to those used in clinic.
Modality and MeasureRelative Gain (%) vs. MI2 Open SourceCXR mAUC5.8XR Exam mAUC2.9MG Cancer AUC7.2CT Lung mAUC9.9US Breast Cancer AUC1.5Fundus mAUC1.9Table 1: Improvements in area under ROC curve compared to open source MI2 according to internal Microsoft research.
Organizations across the healthcare ecosystem can use these proprietary models to help address real clinical needs across diverse healthcare environments.
To help ensure agentic AI delivers meaningful results in healthcare, organizations should select models that have demonstrated effectiveness for their specific tasks. Healthcare AI model evaluator, available now on GitHub, enables teams to test and validate model performance on relevant clinical tasks using their own data and in their own environment. Its intuitive and flexible interface allows organizations to create custom tests and designate human experts or AI models to evaluate the output. Purpose-built for healthcare needs, the evaluator supports evidence-based model selection to help reduce risk and build trust in AI.
Real-world use cases: Impact across diverse healthcare environments
Healthcare agent orchestrator, unveiled at Microsoft Build 2025 and available in the Microsoft Foundry Agent Catalog, features pre-configured agents with multi-agent orchestration and open-source customization options that aid in the creation of agents that coordinate complex workflows. Modular, general reasoners as well as specialized, multimodal AI agents work together to address tasks that could take hours.
Oxford University’s Department of Oncology, in collaboration with Microsoft, has built and deployed three TrustedMDT agents. The agents integrate with Microsoft Teams and summarize patient charts, determine cancer staging, and draft guideline-compliant treatment plans for review by tumor boards. An evaluation and pilot is due to begin at Oxford University Hospitals NHS Foundation Trust in first quarter of 2026. If successful, Oxford hopes to expand into the live clinical pathway following the study’s conclusion, generating vital clinical evidence to support scaling across United Kingdom academic medical centers. Dr. Andrew Soltan MD PhD, National Institute for Health and Care Research (NIHR) Academic Clinical Lecturer and Honorary Specialty Registrar in Medical Oncology at Oxford University Hospitals, shares his story at the Microsoft Ignite breakout session, From Code to Care: Empowering Healthcare with Agentic AI, on Thursday, November 20.
These achievements in agent development showcase the opportunity for impact and reach of agentic AI solutions in diverse healthcare environments—especially when task-specific agents are available for purchase and deployment by other healthcare organizations. In September 2025, Microsoft introduced Microsoft Marketplace, a single destination to find, try, buy, and deploy cloud solutions, AI apps, and agents. Organizations can publish, monetize, and distribute their healthcare agents and solutions through the Marketplace, turning their expertise into scalable business opportunities.
The Atropos Evidence Agent, announced in October 2025 by Atropos Health, answers clinical questions within a physician’s workflow—without the physician even needing to ask a question. Whether it’s for pre-visit planning or during the encounter, the Atropos Evidence Agent draws from patient-level data and surfaces real-world evidence for the provider. Using the healthcare agent orchestrator multi-agent framework, the Atropos Evidence Agent synthesizes real-world data and scientific literature to proactively deliver personalized evidence to clinicians and answer complex questions in minutes without leaving the electronic health record (EHR) workflow. Learn more at their session, Ambient Evidence Automation at the Point of Care, in the Industry Showcase in the Microsoft Ignite Hub on Tuesday, November 18.
UiPath Maestro™ orchestrates specialized Foundry and UiPath agents to synthesize historical records with incidental findings from imaging studies. It transforms raw data into actionable insights that can reduce clinician workload, speed care delivery, and drive measurable impact. Hear their story at the Industry Showcase session, From Insight to Impact: How UiPath Agentic Automation and Azure AI Foundry are Redefining Care Processes, on Thursday, November 20.
From tumor boards workflows to personalized evidence with automated follow-up, these deployments underscore the power of AI for helping to enhance diagnostic speed, accuracy, and efficiency, while supporting safer, more informed care decisions.
Take action to lead the future of healthcare
Microsoft for Healthcare AI is more than a technology upgrade, it’s a practical, high-impact solution for healthcare organizations ready to lead in clinical innovation. With AI models, benchmarking tools, and multi-agent frameworks, leaders can automate workflows, improve patient outcomes, and drive safer care at scale.
The future of healthcare is agentic, data-driven, and collaborative. Take the next step—evaluate, deploy, and scale Microsoft’s AI solutions and work towards advancing care delivery today.
Kees Hertogh
Vice President, Public Sector and Healthcare Marketing at Microsoft
Kees Hertogh serves as vice president of Microsoft’s Public Sector and Healthcare Marketing organization. He leads global product marketing across Healthcare and Life Sciences, Education, and Government, aligning Microsoft’s product portfolios and go-to-market strategies.