Machine Learning Models For Healthcare Predictive Analytics

Today, we're announcing new multimodal models in the MedGemma collection, our most capable open models for health AI development. 𝗪𝗵𝘆 𝗶𝘁 𝗺𝗮𝘁𝘁𝗲𝗿𝘀: Healthcare is increasingly embracing AI. Our Health AI Developer Foundations (HAI-DEF) collection of open models, including MedGemma, provides developers with robust starting points and full control over privacy, infrastructure, and modifications. Today we are announcing two new models: - 𝗠𝗲𝗱𝗚𝗲𝗺𝗺𝗮 𝟮𝟳𝗕 𝗠𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹: Designed for complex multimodal medical reasoning and longitudinal electronic health record interpretation. - 𝗠𝗲𝗱𝗦𝗶𝗴𝗟𝗜𝗣: A lightweight image and text encoder for classification and image retrieval. - All models can be run on a single GPU, with MedGemma 4B and MedSigLIP adaptable to mobile hardware with quantization.  - MedGemma 4B achieves a score of 64.4% on MedQA, state of the art among very small (<8B) open models. - MedGemma 27B models are among the best performing small open models (<50B) on MedQA, scoring 87.7% (text-only). Here’s how researchers and developers around the world have been engaging the MedGemma collection: 🇮🇳Tap Health - exploring MedGemma for its medical grounding, noting its reliability on tasks that require sensitivity to clinical context 🇹🇼 Chang Gung Memorial Hospital - researching how MedGemma works with traditional Chinese-language medical literature and medical staff questions. 🇺🇸DeepHealth - investigating how MedSigLIP could improve their chest X-ray triaging and nodule detection AI efforts. We're eager to see how others in the developer and research community adapt and fine-tune MedGemma! Read the full announcement: https://lnkd.in/dTRJpgng MedGemma technical report: https://lnkd.in/diBR3QTd Explore Health AI Developer Foundations: goo.gle/hai-def Access detailed notebooks on GitHub for inference & fine-tuning;    MedGemma: https://lnkd.in/dFFeMK3g     MedSigLIP: https://lnkd.in/dPpU6kCQ

DeepSeek R1 (mean win rate) and o3-mini (absolute performance) are currently the best performing AI models for healthcare. Based on a recent Stanford benchmark comprised of 121 realistic clinical tasks. Across the board, reasoning models also performed better. 📊 MedHELM benchmark, developed by Suhana Bedi and Nigam Shah (in preprint): 📍 Tested 9 LLMs (GPT-4o, Claude, DeepSeek, Gemini, LLaMA) on ✅ 121 real-world medical tasks ✅ 31 datasets across clinical care, documentation, research, patient comms, admin ✅ 5 categories, 22 subcategories ✅ 29 doctors across 15 specialties validated the taxonomy ✅ Evaluated by a 3-model LLM-jury (had better agreement than human reviewers) 🧪 Examples of what was tested: • “Here’s a patient note — compute the HAS-BLED score.” • “Write ICD-10 codes for this discharge summary.” • “Draft ENT referral based on free-text note.” • “Summarize this patient message in 1 sentence.” 🏆 Best overall performers: • DeepSeek R1 – highest win-rate (66%) • o3-mini – best average accuracy (0.77) • Claude 3.5 Sonnet – nearly tied, at 40% lower cost 🟢 LLMs are strong at: • Writing notes (e.g. discharge summaries): 0.74–0.85 accuracy • Patient messaging + education: 0.76–0.89 🟡 Moderate at: • Clinical decision support (e.g. computing scores, summarizing labs): 0.61–0.76 • Research assistance (e.g. finding evidence, summarizing studies): 0.65–0.75 🔴 Weakest at: • Admin + workflow (e.g. billing codes, referral triage, appointment logic): 0.53–0.63 💡 Takeaways - if you're choosing a model: → Consider reasoning models for better performance → DeepSeek R1 is the strongest overall, especially on documentation and patient messaging → Claude 3.5 has best cost-performance tradeoff → o3-mini leads on decision support accuracy