Art Huston

Big news in healthcare AI—Google just dropped MedGemma and MedSigLIP to the public, and they’re seriously impressive. These models can analyze chest X-rays, medical images, and even patient histories like a digital second opinion. What’s wild is that they don’t just understand text—they can actually see. The larger model, MedGemma 27B, takes in both images and clinical records, and it nailed an 87.7% score on MedQA, which is one of the toughest benchmarks in medical AI. Even the smaller model held its own—U.S. radiologists rated it at 81% accuracy. That’s not just research-grade; it’s actually usable in real patient care. Now here’s the part that hits home for me: I grew up on open-source—Linux distros, GNU apps, the whole ecosystem. So it’s amazing to see Google release these models freely. No paywalls. No cloud-only access. You can download them, tweak them, and run them locally. That’s a huge win for hospitals, researchers, and devs all over the world. It feels like we’re at the start of a new wave—one where powerful AI in healthcare isn’t just for the few with deep pockets, but for anyone who wants to build and improve on it. 🦾 #RoboSapiens #AI #MedTech #ArtificialIntelligence #MachineLearning

344

52 Comments

Here’s your recap of last week’s health IT news 🗞️ 👇 • In case you missed it, we here at Elion raised a $9.3M seed round.   https://lnkd.in/eWa5JEdZ   🤖 𝗔𝗜 𝗖𝗹𝗶𝗻𝗶𝗰𝗶𝗮𝗻 𝗔𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁 • Epic plans to announce its own ambient scribe tool later this month, according to multiple reports.    • Creyos expanded its Session Notes feature to support structured assessment protocols.    • Doximity acquired Pathway Medical in a transaction valued at $63M. 🩻 𝗔𝗜 𝗜𝗺𝗮𝗴𝗶𝗻𝗴 • iRhythm Technologies, Inc. is partnering with Lucem Health to flag elevated arrhythmia risk using clinical data and route identified patients to Zio monitoring—positioned for population health and value-based care programs. 🌡️ 𝗔𝗜 𝗥𝗲𝗺𝗼𝘁𝗲 𝗣𝗮𝘁𝗶𝗲𝗻𝘁 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴 𝗮𝗻𝗱 𝗧𝗿𝗶𝗮𝗴𝗲 • Amwell received a contract extension from the Defense Health Agency to continue powering virtual care for the Military Health System, integrated with MHS GENESIS.    • OU Health partnered with Nexus Bedside to launch a hybrid nursing model that pairs virtual nurses with bedside teams; initial rollout is in neurology/neurosurgery with plans to expand. 💼 𝗣𝗿𝗼𝘃𝗶𝗱𝗲𝗿 𝗮𝗻𝗱 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲 𝗔𝗱𝗺𝗶𝗻𝗶𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻 • Press Ganey launched an integration with UKG that integrates data across the two to automate submissions and inform staffing decisions.    • Capsa Healthcare acquired BlueBin Inc. to “enhance visibility, efficiency, and reliability across the entire clinical supply chain.” 💸 𝗥𝗲𝘃𝗲𝗻𝘂𝗲 𝗖𝘆𝗰𝗹𝗲 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 • Rectangle Health and ECHO Health formed a partnership focused on streamlining payer-to-provider payments for healthcare organizations. 📈 𝗗𝗮𝘁𝗮 & 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 • Rhapsody announced general availability of its Envoy platform for health systems, labs, and payers.    • Duke Health is partnering with Trilliant Health to “inform enterprise strategies in growth planning, network design and patient access.”    • Datavant completed its acquisition of Ontellus. ⚙️ 𝗖𝗹𝗶𝗻𝗶𝗰𝗮𝗹 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀 • Xsolis launched Dragonfly Navigate, an AI-powered workflow tool for discharge planning and capacity management.    • Zyter|TruCare unveiled Zyter Symphony, an agentic orchestration layer to automate enterprise workflows.    • CLEAR is “under construction” in Epic's Toolbox to streamline provider access.    • Cognizant debuted the TriZetto AI Gateway to embed governed generative-AI capabilities across the TriZetto portfolio (claims, care management, customer service).    • Elation Health announced a major AI expansion, adding native AI features (e.g., Note Assist, Actions, Clinical Insights) across its EHR and billing platform.

79

6 Comments

App Ecosystem across EHRs: Patient Care for Epic, Workflow for Athena? After Brendan Keeler posted about the new datasets from the Assistant Secretary for Technology Policy (ASTP), I got tagged and a few people DM'ed me to load these datasets to mimilabs. (Thanks Brendan! ;)) And yes, I loaded them. Now we have data on which hospital/provider uses which EHR, which hospitals are connected to HIE, and what kinds of apps are available in EHR marketplaces. To know more about these datasets, see Brendan's post (https://lnkd.in/eXbfn6va) — he outlined and explained it very well. I also wanted to take a look at the data, but a bit differently from what Brendan found. So here's what I did: I took the EHR app data and decided to look at the distribution of app categories across different EHRs. Since the category names weren't quite standardized, I did a bit of cleaning up. The chart shows the categories of available apps and where those apps are coming from. A few observations: - Epic has all sorts of categories. All possible categories are there. - The #1 category for Epic is Patient Care & Engagement. - The #1 category for athenahealth is Administrative and Workflow. - Since Epic has so many interesting app categories, #2 is misc. - Billing and Revenue, Documentation categories are important for Epic, Athena, as well as other platforms. I wonder if the large driver of Epic having so many patient care apps is due to its client base being hospitals? Doctor-preneurs see more of those needs, maybe? And Athena having more independent practices—they need more help on administrative and workflow to automate with limited resources? What are your thoughts? It's fascinating to see these app ecosystems vary so much across platforms!

79

7 Comments

OpenAI's #OpenSource #GPTOSS was released with strong claims of performance on #healthcare tasks. However, our benchmarks show that it disappoints on benchmarks that weren't built by OpenAI... from medical question answering and reasoning to information extraction and summarization. JSL-MedReasoning-32B, John Snow Labs reasoning model, achieved an average accuracy of 84.99% across public medical benchmarks, surpassing OpenAI’s gpt-oss-120B which scored 77.18%. Both models are optimized for inference on a single 80GB GPU, proving that performance gains don’t require added compute costs. Even JSL-MedS-14B, not a reasoning model, outperformed gpt-oss-20B, and almost matched gpt-oss-120B despite being one-eighth its size. #ai #generativeai #llm #llms #healthcareai #healthai #datascience

58

4 Comments

What topics in AI x healthcare would you be most excited to read about? I’m bringing back my Substack (https://lnkd.in/gu5ay4f) and thinking through topics to dive into around AI × healthcare. Given my background — founder of Avon Health (the first AI-native EMR to receive ONC certification), prior work in ML at Microsoft, PathAI, Carnegie Mellon University, and Stanford University, a patent in cloud ML, and a thesis on computer vision for restoring functional vision in the blind — I’ve been deep in both the technical and business sides of AI in healthcare for almost 15 years. Here are some areas I could explore: - Emerging AI business models in healthcare (pricing, reimbursement, value-based adoption) - Regulatory pathways and ONC/FDA/CMS policy shaping AI adoption - How large language models are transforming clinical workflows - Case studies of AI products that actually made it to production in health systems - Interoperability challenges and how AI can bridge them - The real economics behind AI scribes and automation in care delivery What would you be most excited to read about?

40

6 Comments

Anonymization vs Pseudonymization These two terms are often used interchangeably, but they mean very different things when it comes to protecting patient data. Anonymization is the process of removing or altering personal identifiers so that a person cannot be identified by any means. This process is irreversible. No one can ever link the data back to a person. Pseudonymization  is the process of replacing personal identifiers with artificial identifiers (like codes or random IDs) while keeping a separate key to re-identify the person if needed. It’s reversible if you have the mapping key. e.g “John Smith” becomes “Patient_0098” in the dataset, but a separate file links “Patient_0098” back to “John Smith.” In areas like drug safety, clinical trials, and real-world studies, it’s important to find a balance between protecting privacy and making data useful. Anonymized data works best for open research and sharing results. Pseudonymized data is helpful when you need to follow up with patients, track them over time, or meet regulatory requirements that need re-identification. PS: I have explained many such “confusing but important” concepts in my book Pharma Tech Essentials. Its written to make technology simpler for safety professionals. Available on Amazon: https://lnkd.in/ggRmT5hv

31

1 Comment

Seeing is Knowing: Why Agent Observability is Non-Negotiable in Health & Life Sciences AI As #AIAgents move from pilots to production in #healthcare and #lifesciences, trust, safety, and compliance aren’t optional - they’re mission-critical. Whether you’re a payer optimizing claims, a provider enhancing clinical workflows, or a #pharma/ #MedTech innovator accelerating R&D, the stakes are high: #patientoutcomes, regulatory #compliance, and brand reputation. Enter Agent #Observability: it’s more than monitoring - it’s about deep #visibility into how AI #agents reason, decide, and act across their lifecycle. Why does this matter for our industry? Because in regulated environments, you can’t just know what happened - you need to know why and how. Here are 5 best practices from Microsoft's latest #AgentFactory guidance that every health and life sciences leader should adopt: ✅ Pick the right model with data-driven #benchmarks: Safety, quality, and cost trade-offs matter. Use benchmark-driven leaderboards to select models that meet your compliance and performance needs. ✅ Continuously #evaluate agents in dev & production: Measure intent resolution, task adherence, and response completeness—because a missed step in a clinical workflow isn’t just a bug, it’s a risk. ✅ Integrate evaluations into #CICD: Automated quality and safety checks on every commit ensure your AI evolves without introducing compliance gaps. ✅ #RedTeam before go-live: Simulate adversarial scenarios to uncover vulnerabilities before they impact patient care or data integrity. ✅ #Govern with confidence: Embed policies aligned with frameworks like the EU AI Act and HIPAA to maintain trust and accountability. Bottom line: In health and life sciences, responsible AI isn’t a checkbox - it’s a continuous discipline. Agent observability is your foundation for safe, scalable, and compliant AI innovation. 👉 How is your organization approaching AI observability? Are you ready to make it part of your compliance and quality strategy? Learn more here: https://lnkd.in/e7SZjx6B #AI #HealthcareInnovation #LifeSciences #ResponsibleAI #AzureAI #AgenticAI #HealthcareAI #HealthTech #AIObservability

26

3 Comments

Sunday hot take: We won't see AI impact clinical care until we have more physicians as builders. The debate on technical vs non-technical or how technical shouldn't build without clinical is so tired. Because right now the entire focus should be on how to make non-technical CLINICAL technical. And no, not technical in the sense of retraining or understanding how to code. But technical in the sense that they pair program with AI directly within clinical applications and care delivery environments. EVERYTHING meant for clinical should be built this way, from scientists in labs, to doctors in rural health care settings. Builder first, not just user or buyers. ------- Hi, I'm Dr. Mario Amaro, Founder and CEO at Cline and we're working on something new that will finally help empower doctors to transition from being "just buyers" to becoming BUILDERS. Follow us to learn more. #HealthcareOnLinkedIn

52

13 Comments

One of the most underrated skills in healthcare AI? Learning how your clients work before showing them how your product works. Every healthcare leader we speak with has their own way of doing things — different processes, documentation standards, risk adjustment models, EHR workflows. No two teams operate alike. That’s why at CombineHealth, we're not aiming to be just another vendor. We aim to be a true partner — one that listens first, configures second, and scales only when value is clear. We adapt our AI agents to each customer’s workflow — whether that means integrating with their systems, customizing logic to meet internal policies, or flagging only the cases where human oversight matters. Because in healthcare, trust isn’t built on shiny features. It’s built on alignment.

27

2 Comments