Collaborative AI Models for Medical Diagnosis

Can an # AI #Doctor partner with clinicians? Can we please move past the AI versus doctor/clinician comparisons in taking board exams.. solving diagnostically challenging cases... providing more empathetic on-line responses to patients...? and instead focus on improving patient care and their outcomes? The authors, Hashim Hayat, Adam Oskowitz et. al. at the University of California, San Francisco, of a recent study may be hinting at this: envisioning an agentic model (Doctronic) “used in sequence with a clinician” to expand access while letting doctors focus on high‑touch, high‑complexity care and supporting the notion that AI’s “main utility is augmenting throughput” rather than replacing clinicians (https://lnkd.in/e-y3CnuF)  In their study: ▪️ >100 cooperating LLM agents handled history evaluation, differential diagnosis, and plan development autonomously. ▪️ Performance was assessed with predefined LLM‑judge prompts plus human review. ▪️ Primary diagnosis matched clinicians in 81 % of cases and ≥1 of the top‑4 matched in 95 %—with no fabricated diagnoses or treatments. ▪️AI and clinicians produced clinically compatible care plans in 99.2 % of cases (496 / 500).  ▪️In discordant outputs, expert reviewers judged the AI superior 36 % of the time vs. 9 % for clinicians (remainder equivalent). Some key #healthcare AI concepts to consider: 🟢 Cognitive back‑up, in this study, the model identified overlooked guideline details (seen in the 36 % of discordant cases; the model used guidelines and clinicians missed). 🟢 Clinicians sense nuances that AI cannot perceive (like body‑language, social determinants). 🟢 Workflow relief , Automating history‑taking and structured documentation, which this study demonstrates is feasible, returns precious time to bedside interactions. 🟢 Safety net through complementary error profiles – Humans misdiagnose for different reasons than #LLMs; so using both enables cross‑checks that neither party could execute alone and may have a synergistic effect. Future research would benefit from designing trials that directly quantify team performance (clinician/team alone vs. clinician/team + AI) rather than head‑to‑head contests, aligning study structure with the real clinical objective—better outcomes through collaboration. Ryan McAdams, MD Scott J. Campbell MD, MPH George Ferzli, MD, MBOE, EMBA Brynne Sullivan Ameena Husain, DO Alvaro Moreira Kristyn Beam Spencer Dorn Hansa Bhargava MD Michael Posencheg Bimal Desai MD, MBI, FAAP, FAMIA Jeffrey Glasheen, MD Thoughts? #UsingWhatWeHaveBetter

This blog highlights the launch and significance of Microsoft’s Healthcare Agent Orchestrator, a powerful AI-driven platform designed to support complex, multidisciplinary medical workflows—most notably in cancer care. Key Significance:    •   Cancer treatment is highly personalized, but <1% of patients currently benefit from fully personalized care plans due to the high cost, time, and effort involved.    •   Multimodal Agentic AI can dramatically reduce the hours clinicians spend on reviewing complex patient data.    •   Microsoft’s platform enables orchestrated collaboration among specialized AI agents to streamline these workflows and integrate into tools clinicians already use (e.g., Microsoft Teams, Word, Copilot).    •   The goal is to scale precision medicine, speed up decision-making, and augment—rather than replace—human experts. Examples of Specialized Agents: 1. Patient History Agent – Builds a chronological patient timeline using Universal Medical Abstraction. 2. Radiology Agent – Provides a “second read” of medical imaging, using models like CXRReportGen/MAIRA-2. 3. Pathology Agent – Can link with external pathology agents like Paige.ai’s Alba, analyzing tumor slides. 4. Cancer Staging Agent – Applies AJCC clinical guidelines to accurately determine cancer stages. 5. Clinical Guidelines Agent – Uses NCCN guidelines to recommend treatments. 6. Clinical Trials Agent – Matches patients to trials, improving recall over baseline models. 7. Medical Research Agent – Synthesizes research findings into actionable clinical insights. 8. Report Creation Agent – Generates integrated, formatted reports for tumor boards. Real-World Impact & Collaborators:    •   Stanford Health Care, Johns Hopkins, UW Health, Mass General Brigham, and Providence Genomics are actively piloting or integrating these agents.    •   Real use cases include enhancing tumor board meetings, streamlining clinical trial matching, and deepening pathology insight via conversational interfaces (e.g., Paige.ai’s Alba in preview). Bottom Line: The healthcare agent orchestrator marks a pivotal step in democratizing precision oncology, accelerating collaboration between AI and human experts, and scaling care excellence through modular, customizable AI agents. It’s already in the hands of top institutions and could revolutionize how we approach cancer treatment at scale.

🚀 AI Agent-Powered Multi-Medical Diagnostics 🔍 Redefining Diagnosis Through Multi-Modal Intelligence and Adaptive Clinical Reasoning 🧠 As the complexity of patient presentations increases—with comorbidities, fragmented data, and diagnostic uncertainty—traditional clinical models are being pushed to their limits. ✨ AI agents: autonomous, modular, learning-capable systems that serve as real-time Clinical Decision Support Systems (CDSS). These agents don’t replace clinicians—they augment their reasoning by integrating imaging, labs, genomics, wearables, and clinical notes into a unified, intelligent diagnostic interface. 📊 From multi-modal data fusion to multi-diagnostic reasoning, AI agents tackle diagnostic silos and provide contextualized insights at the point of care. Whether it’s parsing symptoms of cardiac distress across CT, ECG, and labs—or simulating comorbidity scenarios like sepsis vs. hepatic encephalopathy—they help clinicians think deeper, faster, and more accurately. 🧬 Powered by a modular architecture (perception, cognition, memory, world models, and emotion-aware interaction), these agents represent a leap in diagnostic intelligence—yet always under human oversight. 🔮 While widespread clinical deployment is still on the horizon, the vision is compelling: 🔗 Unified diagnostic assistants in radiology and oncology 🩺 Ambient monitoring for cardiopulmonary deterioration 🧑⚕️ AI copilots in primary care visits 🌍 Equitable diagnostics through decentralized AI agent networks ⚖️ But with innovation comes responsibility: explainability, trust, regulation (FDA SaMD), and ethical design are key to success. 💡 The future of diagnostics isn’t just automated. It’s collaborative, context-aware, and AI agent-augmented—designed to elevate human clinical judgment, not replace it. #AIinHealthcare #Diagnostics #ClinicalDecisionSupport #MultiModalA #HealthTech #MedicalAI #DigitalHealth #AIagents

AI just ran its own multidisciplinary tumor board. And nailed the diagnosis + treatment. This was a full-stack oncology reasoning engine—pulling from imaging, pathology, genomics, guidelines, and literature in real time. A new paper in Nature Cancer describes how researchers built a GPT-4-powered multitool agent that: • Interprets CT & MRI scans with MedSAM • Identifies KRAS, BRAF, MSI status from histology • Calculates tumor growth over time • Searches PubMed + OncoKB • And synthesizes everything into a cited, evidence-based treatment plan In short: it acts like a multidisciplinary team. Results : • Accuracy jumped from 30% (GPT-4 alone) to 87% • Correct treatment plans in 91% of complex cases • Every conclusion backed by a verifiable citation This is bigger than oncology. Any field that relies on multi-modal data and cross-domain reasoning—like my field of GI ( GI + Mental Health+ Nutrition + Excercise ) could benefit from this collaborative AI architecture. Despite the visual, it doesn’t replace the human team—it augments it. Providers still decide. But now, they do it faster, with more context, and less cognitive fatigue. #AI #HealthcareonLinkedin #Healthcare #Cancer

Imagine AI models working together like a team of doctors, each contributing their expertise to solve complex medical cases. This is what "MDAgents: An Adaptive Collaboration of LLMs for Medical Decision-Making" explores, as recently presented at NeurIPS 2024. Working: MDAgents brings a novel approach to using Large Language Models (LLMs) in medicine by dynamically creating a collaborative environment tailored to the complexity of each medical query: 1) Complexity check: Each medical question is evaluated for complexity, determining whether it necessitates a basic, moderate, or advanced collaborative response. 2) Expert recruitment: Based on complexity, MDAgents "recruits" AI agents to act as specialists, forming either a solo practitioner model, a Multi-disciplinary Team (MDT), or an Integrated Care Team (ICT). 3) Analysis and synthesis: The agents engage in collaborative reasoning, using techniques like Chain-of-Thought (CoT) prompting to draw insights and resolve disagreements for more nuanced cases. 4) Decision-making: Synthesizing diverse inputs, the framework reaches a final decision, informed by external medical knowledge and structured discussions among the AI agents. Achievements: 1) MDAgents outperformed both solo and group LLM setups in 7 out of 10 medical tasks, enhancing decision accuracy by up to 11.8%. 2) Demonstrated the critical balance between performance and computational efficiency by adapting the number of participating agents based on task demands. Link to the full paper -> https://lnkd.in/gR7Zwm7t #AI #Healthcare #NeurIPS2024 #MedicalAI #Collaboration #InnovationInMedicine #ResearchInsights

Medical AI Just Took a Giant Leap Forward Microsoft's latest research is reshaping what's possible in healthcare AI. Their MAI Diagnostic Orchestrator (MAI-DxO) achieved 85% diagnostic accuracy on complex New England Journal of Medicine cases—four times better than experienced physicians. But here's what makes this breakthrough different from typical AI hype: 🧠 It thinks like a medical team, not a search engine Instead of building a single super-intelligence, Microsoft built a team—a digital department of AI doctors, each playing a role in a simulated clinical environment. One AI generates hypotheses, another challenges them, and a third manages cost-effectiveness. 💡 It mimics real clinical reasoning Rather than relying on multiple-choice questions that favor memorization, the system uses sequential diagnosis—starting with symptoms, asking questions, ordering tests, and iteratively refining the diagnosis. Just like how doctors actually work. 📊 The efficiency gains are remarkable The AI reduces diagnostic costs by 20% compared to human doctors while maintaining higher accuracy. For B2B healthcare tech, this represents the kind of ROI that drives real adoption. From my experience building AI-powered solutions, what excites me most is the orchestration approach. This isn't about replacing doctors—it's about creating AI systems that collaborate like expert teams, something we're seeing across many industries. The broader implication? We're moving from AI that answers questions to AI that reasons through complex problems. This pattern will reshape not just healthcare, but how we approach problem-solving in cybersecurity, customer support, and beyond. Read more: https://lnkd.in/gjNU64yP P.S. - With over 50 million health-related searches happening daily across Microsoft's AI products, this research has immediate practical applications.

Microsoft just tested its new diagnostic AI on 304 of medicine’s hardest cases: based on real patient records from MGH. AI diagnosed correctly 4x more than doctors (study in preprint). Led by Eric Horvitz, Harsha Nori and team. 📌 What was done • Converted NEJM cases into stepwise challenges: AI or doctors start with an initial presentation, iteratively ask questions, order tests, and diagnose • Benchmarked leading AI models (GPT, Claude, Gemini, Llama, Grok, DeepSeek) alongside 21 US & UK physicians (5–20 years’ experience) • Built MAI-DxO, an orchestrator coordinating virtual AI “doctors” to track hypotheses, weigh uncertainty, and order cost-effective tests • Doctors solved cases via text-chat with a Gatekeeper model revealing info only when asked 📌 Key results • MAI-DxO: 85% accuracy (NEJM cases) • Physicians: 20% accuracy • MAI-DxO achieved accuracy with fewer, targeted tests, lowering diagnostic costs compared to doctors or baseline AI AI models (accuracy / estimated diagnostic cost per patient): • MAI-DxO (o3): 85% / ~$7,200 • OpenAI o3: 79% / ~$7,850 • Gemini 2.5 Pro: 69% / ~$4,800 • Claude Sonnet/Opus: 60–70% / ~$6,000–7,000 • Llama/DeepSeek: 40–55% / ~$4,000–5,000 • GPT-4o: 50% / ~$2,750 • Human physicians: 19.9% / ~$2,963  ⚠️ Limitations • Physicians were not allowed to use external references or the internet • Participants were generalists (17 primary care, 4 hospital medicine), not specialists 🔷 Why it matters • Cost analysis shows AI didn’t “win” by ordering every possible test • With Microsoft keen on integrating AI into Bing/Copilot, this work could accelerate how diagnostic reasoning features in consumer health search, and perhaps improve first-line health advice for millions I was a bit surprised that to "create a fair comparison with the AI", doctors were not allowed to use the internet or references like UpToDate (H/T Maxime Griot, MD). Means severe underestimation of real-world doctor performance, especially when participants were reviewing difficult NEJM cases outside their specialty. Will leave it to readers to judge how well the study addresses other limitations. Jerome Kassirer, MD (former NEJM editor-in-chief) has a simple checklist for assessing "AI vs doctor" studies: → Cases real or made up? → Full of messy ambiguity clinicians face? → Enough cases tested? → Were diagnoses evidence-based or subjective? → Who judged accuracy: were humans and AI given the same information? → Were disagreements fairly adjudicated? For me, the standout strength of this study was cost analysis. Unnecessary testing suggested by AI will drive up costs without improving care or outcomes. Seeing explicit cost-benefit analysis in an AI study is rare, and important. Also notable: much of this work was driven by AI and clinical researchers based in London (eg. Christopher Kelly, Viknesh Sounderajah, Xiao Liu). The UK has great AI expertise. #AIinHealthcare #DiagnosticAI #ClinicalAI #HealthTech #MedicalAI #MicrosoftAI

Microsoft’s new AI outperformed real doctors - on real, published medical cases. It's called the Microsoft AI Diagnostic Orchestrator (MAI-DxO). And it was tested against 21 clinicians in complex cases. The results: MAI-DxO = 85.5% accuracy Medical doctors = 20% accuracy What's driving these results: ▶︎ 1. Systematic validation approach Rather than cherry-picking cases, Microsoft created a standardized benchmark using real-world scenarios where the outcomes were already known. ▶︎ 2. Tackling the hardest cases MAI-DxO was tested on 304 case reports from the New England Journal of Medicine - the kind that usually stump individual doctors and require specialist teams. It consistently made more accurate calls, with fewer diagnostic tests. ▶︎ 3. Multiple AI "specialists" working together The system doesn't rely on a single AI model. Instead, it orchestrates multiple agents - each acting like a domain expert - to simulate how medical teams collaborate on diagnoses. ▶︎ 4. Built-in cost consciousness Every test ordered had a virtual price tag. The AI had to weigh diagnostic value against cost - just like real doctors managing limited resources and insurance constraints. But here's the reality they aren’t being vocal about: → The doctors in this study couldn't use reference tools like UpToDate (the medical database most physicians rely on) or search engines. → The AI hasn't been tested in actual clinical settings yet. → And regulatory approval is still pending. Still, this represents a significant shift in how we should think about AI in healthcare. We're moving from "AI as a helpful assistant" to "AI as a diagnostic partner" - one that might soon outperform human expertise in complex cases. The question isn't whether this technology will transform healthcare. It's how quickly healthcare systems will adapt to integrate it. What's your take on AI becoming better than doctors at diagnosis? #entrepreneurship #AI #healthtech