Enhancing Diagnostics with Machine Learning

Exciting Innovation in Healthcare AI: MedRAG I just came across a groundbreaking paper introducing MedRAG - a novel approach that enhances Retrieval-Augmented Generation (RAG) with Knowledge Graph-Elicited Reasoning for healthcare applications! Diagnostic errors are a serious problem in healthcare, with approximately 795,000 individuals suffering permanent disability or death annually due to misdiagnosis in the US alone. MedRAG addresses this challenge by significantly improving the accuracy and specificity of AI-powered diagnostic support. >> How MedRAG Works: The system combines RAG with a comprehensive four-tier hierarchical diagnostic knowledge graph to enhance reasoning capabilities. Here's the technical breakdown: 1. Diagnostic Knowledge Graph Construction: MedRAG systematically builds a four-tier hierarchical diagnostic KG through disease clustering, hierarchical aggregation, and LLM augmentation. This captures critical diagnostic differences between diseases with similar manifestations. 2. Diagnostic Differences KG Searching: When a patient's manifestations are input, the system decomposes them into clinical features, embeds them, and matches them with relevant diagnostic differences through multi-level matching and upward traversal within the KG. 3. KG-elicited Reasoning RAG: The system retrieves relevant Electronic Health Records (EHRs) and integrates them with the identified diagnostic differences KG to trigger reasoning in a large language model, generating precise diagnoses and treatment recommendations. 4. Proactive Diagnostic Questioning: MedRAG can identify when patient information is insufficient and proactively suggest follow-up questions based on discriminability scores of features in the knowledge graph. The researchers evaluated MedRAG on both a public dataset (DDXPlus) and a private chronic pain diagnostic dataset from Tan Tock Seng Hospital. It outperformed state-of-the-art RAG models, achieving up to 11.32% improvement in diagnostic accuracy for diseases with similar manifestations. What's particularly impressive is MedRAG's compatibility across various backbone LLMs, including open-source models like Mixtral-8x7B and Llama-3.1-Instruct, as well as closed-source models like GPT-4o. This technology has tremendous potential to reduce misdiagnosis rates and improve healthcare outcomes by providing more accurate, specific diagnostic support and personalized treatment recommendations.

We’re excited to share research from our health team at Microsoft AI: a proof-of-concept showing that AI can master medicine’s most intricate diagnostic challenges by following the same step-by-step reasoning expert physicians use. There's more detail in our pre print paper & blog Paper-> https://lnkd.in/egDiNsqR Blog-> https://lnkd.in/esGFhSeB Sharing what I'm most excited about from this work. 1. Benchmarks Traditional medical benchmarks like the USMLEs condense clinical cases into neat multiple-choice questions—far from the real clinical workflow. We’ve approached things in a different way: Sequential Diagnosis Benchmark (SDBench) deconstructs 304 of the most diagnostically complex and demanding cases in medicine published in the New England Journal of Medicine. SD Bench requires models—and physician—to begin with an initial presentation, ask follow-up questions, order tests, and converge on the confirmed diagnosis—just as in routine clinical practice. You can see how this works in a video with Xiao Liu on our blog. 2. Performance With this new benchmark we tested a suite of the best known generative AI models against the 304 NEJM cases with impressive out-the-box performance. Beyond this we developed the Microsoft AI Diagnostic Orchestrator (MAI-DxO). By emulating a virtual panel of physicians with diverse thinking styles, MAI-DxO boosts raw model accuracy and solves a remarkable 85.5% of NEJM cases. For comparison we evaluated 21 practicing UK/US physicians and on the same tasks, these experts achieved a mean accuracy of 20%. 3. Costs One of our concerns was that AI would default to ordering every investigation to arrive at the correct diagnosis. So we set the system up such that each requested investigation also incurred a cost. This allowed us to evaluate performance against both diagnostic accuracy and resource expenditure. As MAI-DxO is configurable it is seen to operate along a Pareto frontier of accuracy versus resource use. What’s next: While for now exciting research, we believe this kind of superhuman clinical reasoning will in future reshape medicine. A particular focus for our group is on consumer health. Today, Bing and Copilot answer over 50 million health queries daily—from a first-time knee-pain search to finding a late-night pharmacy. We’re committed to bringing rigorous and reliable AI support into these journeys, backed by clinical evidence and robust commitments to quality, safety and trust. A huge shout-out to everyone on our new team who contributed, our partners across Microsoft, and particularly to Mustafa Suleyman for his vision. He saw the opportunity for AI to improve healthcare more than a decade ago and it now feels like this is the right time to deliver on the opportunity. Harsha Nori Mayank Daswani Christopher Kelly Scott Lundberg Marco Túlio Ribeiro Marc Wilson Xiao Liu Viknesh Sounderajah Bay Gross Peter Hames Eric Horvitz Charlotte Cooper Simpson, PhD

How good is generative AI at diagnosis compared to human doctors? Today, we introduce a novel, interactive medical benchmark (SDBench) for “sequential diagnosis”, and an orchestrator (MAI-DxO) that achieved over 4x higher diagnostic accuracy vs experienced physicians who played the benchmark. 📈 Introducing SD Bench: * 304 New England Journal of Medicine case reports, re‑imagined as an interactive generative AI game, playable by both humans and AI agents. * Ask any question, order any test - SD Bench replies with real or high-fidelity, coherent, simulated findings based upon the case. * Two fundamental metrics: Did they get it right? And what did it cost? 🤖 Enter MAI-DxO – our model-agnostic orchestrator * Simulates a multidisciplinary panel of doctors, with carefully chosen roles, debating and selecting the next best diagnostic question or test to order. * Plugs into language models from any frontier lab—OpenAI, Anthropic, Google, xAI, DeepSeek, Meta... Same orchestration logic lifts all models - proving that structured, role-based reasoning scales. * When using OpenAI’s o3, it achieved 80% accuracy vs 20% for UK/US physicians (4x higher!), while spending 20% less on diagnostic tests. When configured for maximum accuracy, MAI-DxO achieves 85.5% accuracy. 💡 Why it matters AI shouldn’t just know the answer to a single question; it must ask the right questions at the right time, adjust given new information, weigh up costs/invasiveness of investigations, and collaborate deeply like the best physicians. While just a game at this stage, SDBench + MAI-DxO feels like an important step towards that future. Preprint: https://lnkd.in/eETbR7ku Blog: https://lnkd.in/ep4wuhB8 Ps we plan to release the full benchmark soon. Stay tuned. Subscribe to this thread for updates! 👀 Congrats to Harsha Nori, Mayank Daswani, Christopher Kelly, Scott Lundberg, Marco Túlio Ribeiro, Marc Wilson, Xiao Liu, Viknesh Sounderajah, Jonathan Carlson, Matthew Lungren MD MPH, Bay Gross, Peter Hames, Mustafa Suleyman, Dominic King, MD PhD, Eric Horvitz 🥳

Google DeepMind and Stanford just compressed a 5-7 year diagnostic odyssey into actionable insights. Their AI correctly identified causative genes for rare diseases - including a novel mutation for hearing loss that was later validated in the lab. Published in 𝘈𝘥𝘷𝘢𝘯𝘤𝘦𝘥 𝘚𝘤𝘪𝘦𝘯𝘤𝘦, this isn't just another AI research paper. It's a blueprint for fundamentally changing how rare disease patients get diagnosed. 𝗧𝗵𝗲 𝗕𝗿𝗲𝗮𝗸𝘁𝗵𝗿𝗼𝘂𝗴𝗵 Researchers at Google DeepMind and Stanford University demonstrated that large language models can dramatically accelerate rare genetic disease diagnosis. Using Google's Med-PaLM 2 and Gemini 2.5 Pro, the team analyzed complex genetic and clinical data to identify causative genetic factors in both mouse models and human patients. 𝗪𝗵𝗮𝘁 𝗠𝗮𝗸𝗲𝘀 𝗧𝗵𝗶𝘀 𝗦𝗶𝗴𝗻𝗶𝗳𝗶𝗰𝗮𝗻𝘁 The AI solved genetic problems of increasing complexity with remarkable precision: • Identified a novel causative gene for hearing loss in mice (later validated in the lab) • Analyzed genomic data from human patients with multifaceted symptom profiles • Successfully pinpointed underlying genetic variants, including variants of unknown significance (VUS) The system uses a retrieval and grounding pipeline to analyze vast amounts of genetic information and generate ranked hypotheses - essentially reasoning through genetic data the way a skilled clinical geneticist would, but at scale. 𝗜𝗺𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀 𝗳𝗼𝗿 𝗣𝗵𝗮𝗿𝗺𝗮 & 𝗛𝗲𝗮𝗹𝘁𝗵𝗰𝗮𝗿𝗲 For rare disease drug development, faster diagnosis means: • Earlier patient identification for clinical trials • More accurate patient stratification • Accelerated pathway from genomic discovery to targeted therapy development • Reduced diagnostic odyssey costs (currently averaging $5M per patient over their journey) This represents more than incremental progress in AI-assisted diagnostics. It's a fundamental shift in how we might approach precision medicine - compressing years of diagnostic uncertainty into actionable insights that enable faster therapeutic intervention. The question isn't whether AI will transform rare disease diagnosis. It's how quickly we can validate and implement these tools in clinical practice. Follow Dr. Suzanne Morgan for more insights on AI and Rare Disease Source: https://lnkd.in/d-ki3nNu

🧪 New Machine Learning Research: Enhancing 3D Medical Image Annotation with SAM Integration Zafer Yildiz and his colleagues from Duke University have introduced a new extension for 3D Slicer, integrating the Segment Anything Model (SAM) and SAM 2 to improve 3D medical image annotation. - Research goal: To enhance the efficiency and accuracy of 3D medical image annotation by adapting SAM and SAM 2 models within the 3D Slicer software. - Research methodology: The team integrated SAM and SAM 2 into 3D Slicer, enabling both single-directional and bi-directional propagation of segmentation masks across 3D volumes. The extension allows users to interactively place point prompts on 2D slices, which are then used to generate and propagate annotations throughout the entire 3D volume. - Key findings: The adapted SAM 2 model demonstrated effective segmentation across different medical imaging modalities, including CT and MRI. For example, it achieved a 33% increase in segmentation accuracy when using bi-directional propagation compared to traditional methods. - Practical implications: This extension reduces the time required for medical image annotation by up to 50%, providing a valuable tool for radiologists and medical professionals in applications like tumor detection, organ segmentation, and pre-surgical planning. Stay tuned for more updates on the latest advancements in ML and data science! #LabelYourData #MachineLearning #Innovation #AIResearch #MLResearch #MedicalImaging #DataScience #MedicalAI

Earlier this year, a close family member was dangerously ill in India. The diagnosis wasn’t working. The symptoms were escalating. No one knew why. We felt that familiar dread - being far from the situation, and even farther from certainty. So I did what millions of people now do in moments of uncertainty: I asked ChatGPT. Typed in the symptoms, context, and history - not expecting magic, just hoping for perspective. What came back was startlingly precise: It could be this. If so, check the kidneys. If the kidneys are involved, watch for infection. If it’s in the blood, it could be sepsis. Escalate - fast. It was right. All of it. We flagged it to the doctors. It shaped the next set of tests. And it helped turn a very bad situation around- fast. That moment crystallized something for me: AI isn’t about replacing doctors. It’s about replacing helplessness. There’s a lot of talk in Silicon Valley about curing death. Like, literally - curing aging, reversing entropy, building new bodies from cells that forgot they were old. Some of it will work. Much of it will take decades. But the more immediate, life-changing breakthroughs are already happening - not at the edge of life, but at the frontlines of medicine. Just this week: ▪️Microsoft AI Diagnostic Orchestrator (MAI‑DxO) outperformed experienced physicians. It was tested on 304 real-world case studies published in The New England Journal of Medicine. MAI-DxO solved 85.5% of them. By comparison, 21 experienced physicians solved just 20%. How? By mimicking a panel of clinical minds: One AI model orders tests, another evaluates the results, others debate, reframe, escalate. It’s a structured, chain-of-thought system modeled on real diagnostic reasoning. And it recommended fewer unnecessary tests, saving both time and cost. Yes, it’s early. It hasn’t been deployed in hospitals. But the signal is loud: we’re not far from AI-powered co-pilots for frontline care. ▪️ Google DeepMind's AlphaGenome, tackled a different frontier: the "dark matter" of DNA. Most disease-causing mutations don’t lie in genes, they hide in the regulatory code. Until now, we couldn’t see them at scale. AlphaGenome can process 1m base pairs at once - entire genomic neighborhoods. It’s already predicted how some non-coding mutations can trigger cancer. And it trained in just four hours. If MAI‑DxO gives us a better map of what’s happening now, AlphaGenome gives us a telescope into what might happen next. These tools don’t just answer questions. They reshape who gets to ask them. This is what makes the AI revolution in medicine so powerful. Not just that it might one day extend life. But that it already extends understanding. That it makes complexity legible. That it turns patients into partners - and doctors into augmented super-thinkers. And that alone could save millions of lives.

Microsoft’s research team just tested an AI diagnostic tool on over 100 difficult cases. The system—MAI-DxO—reached the correct diagnosis 85% of the time. General practitioners hit 20%. This isn’t a future scenario. It’s a working prototype. It asks questions, reviews tests, revises its thinking. Like a doctor would—only faster and with more recall. It’s not cleared for clinical use yet. But it’s showing how generative AI might support real diagnostic reasoning, not just spit out search results. What matters here isn’t just the accuracy. It’s the process. The AI shows its reasoning. It collaborates. It invites review. That’s closer to human judgment than most tools we’ve seen. If it holds up, it could reduce error, lower costs, and speed up decision-making in overburdened systems. But it also raises questions: Who reviews the output? Who’s liable? How do we train people to work with a tool that can outperform them? This is less about replacing doctors and more about rethinking the role. And building systems that improve care without removing care. Article: https://lnkd.in/eDz4rq3U