Cassi Ericson’s Post

In 2019, I had a health scare that involved a series of misdiagnoses, surgery, and a $100,000 bill. My situation isn't uncommon. Med-Gemini, by Google, offers a glimpse of better healthcare outcomes👇 It's fascinating to see what Med-Gemini is capable of and the implications it could have for patients as a whole and personally. They used the base Gemini and fine-tuned it for medical tasks that involve diagnosis. → It has multimodality capability as it can read and diagnose 2D (e.g. X-rays) and 3D (e.g. CT Scan) to generate commentaries about the patient's condition. → It involved fine-tuning and evaluating across 14 different data sources ranging from chest X-ray data (MIMIC-CXR), skin-lesion images (PAD-UFES-20), to lung CT scans (NLST). → It achieved 91.1% accuracy on a popular MedQA benchmark (a collection of medical board exams), beating the benchmark of other SOTA's like GPT-4 by 0.9%. Think of an array of benefits that AI could bring to medicine. The implications are profound and wide. → Accurate diagnosis - About 7.4M misdiagnoses happen in the U.S. every year. With the help of AI, errors in diagnosis could be greatly reduced. → Cost reduction - The total cost of healthcare expenses in the U.S. every year is a whopping $4.5T (or about ~13.5K per person). AI diagnostics could greatly reduce visits and tests, which would save costs. → Comprehensive care - Imagine a 360-degree overview of your personal health. AI could provide more coverage on health conditions and risks based on your medical data. Early prevention is the big thing here. 👉 What do you think about the application of AI in healthcare? Do you think it will benefit patients? Drop a comment👇 👉 Smash 👍 and follow Daniel Lee to land dream data & AI jobs 👉 Break into Data Science and ML Engineering with datainterview.com 🚀

Efficient care starts with smart predictions. Machine Learning drives smarter #healthcare by analyzing #patientdata to optimize surgery schedules, admissions and emergency triaging, and to improve care delivery and reduce delays. Via https://lnkd.in/ecw3Pkmv #machinelearning #artificialintelligence #ai #healthcareinnovation 

AI does Not Necessarily Lead to more Efficiency in Clinical Practice Although AI is often seen as a solution for handling routine tasks such as monitoring patients, documenting care tasks and supporting clinical decisions, the actual effects on work processes are unclear. Particularly in data-intensive specialties such as genomics, pathology and radiology, where AI is already being used to recognise patterns in large amounts of data and prioritise cases, there is a lack of reliable data on efficiency gains. The results of this analysis makes clear that the use of AI in everyday clinical practice that local conditions and individual work processes have a major influence on the success of AI implementation. #AI #healthcare #workflows #efficiency #improvement https://lnkd.in/eB6SDgnZ

Harrison.ai just launched the most performance radiology specific vision language model (VLLM) to date scoring 51.4 out of 60 (85.67%) on the Fellowship of the Royal College of Radiologists (FRCR) Rapid 2B exam 2x the score of other frontier models from OpenAI, Microsoft, Google and Anthropic. Harrison.rad.1 isn't just another AI—it's a specialised medical model with an open-ended dialogue interface that can detect and localise radiological findings, generate clinical-grade structured reports provide longitudinal reasoning based on patient history and more. It's the only multimodal foundational model in imaging trained on a proprietary high-quality dataset of millions of longitudinal radiology studies and instruction finetuned at an unprecedented scale by our team of hundreds of medical specialists with supervised annotation. Factual correctness and clinical precision are the top priority in the model development. I believe AI has incredible potential to support medical practitioners, especially with the ever-growing volume of radiological images and the global shortage of medical professionals. We're making this model accessible to researchers, healthcare professionals, and regulators because we believe in sparking a collective conversation on how we can innovate healthcare using AI—safely, ethically, and responsibly. Everything we do is about using AI responsibly to improve healthcare, visit our website if you would like to join us in this mission of developing responsible AI. Join the waitlist for early access to the model: https://lnkd.in/g8t_eZzT

**Exciting Breakthrough in Radiology AI!** Harrison.ai's launch of *Harrison.rad.1*, a cutting-edge radiology-specific Vision Language Model (VLLM), marks a monumental step forward for healthcare. Scoring 85.67% on the FRCR Rapid 2B exam—doubling the performance of frontier models from industry giants like OpenAI and Google—this model is more than just a technological advancement. It's a specialized AI tool that can transform radiological diagnostics, providing unparalleled accuracy and efficiency. With its ability to detect, localize findings, generate structured reports, and apply longitudinal reasoning, *Harrison.rad.1* is poised to tackle the growing volume of medical imaging and the shortage of professionals. The fact that it’s trained on millions of radiology studies and fine-tuned by hundreds of medical specialists shows the commitment to factual accuracy and clinical precision. Kudos to Harrison.ai for making this model accessible to researchers and healthcare professionals alike. This is a prime example of how responsible, ethical AI development can significantly advance healthcare, paving the way for better patient outcomes. I can't wait to see how this innovation shapes the future of radiology!

Advancing medical AI with Med-Gemini Google's Med-Gemini, a new AI model family adapted for healthcare, shows strong potential in diverse medical tasks through advanced reasoning and multimodal understanding. Research papers highlight Med-Gemini's success on benchmarks and its unique capabilities in interpreting complex medical data, answering clinical questions, and generating reports.evaluations. Notably, it demonstrates the model's ability to interpret complex 3D scans, answer clinical questions, generate radiology reports, and encode genomic information for risk prediction with strong results across various diseases. #GenAI #GoogleCloud #Gemini #GoogleAI

https://tcrn.ch/3xszfxx Interesting piece in TechCrunch about what #generative AI could mean for healthcare. Some key points: ⭐ The article described the potential of using generative AI in #radiology. ⭐ Worth noting that both examples given in the radiology section aren't actually #generativeAI though (both examples are machine learning). ⭐ Despite enthusiasm from the investor community, patients and clinicians are less sure about its value ⭐ Only 53% of surveyed consumers thought it could add value in healthcare ⭐ A paper in JAMA Pediatrics found ChatGPT made errors diagnosing pediatric diseases 83% of the time Despite reservations, I think there's no doubt that generative AI could be game-changer, especially for routine, mundane tasks. Be interesting to hear what others think ! #radiology #AI

A.I. is advancing rapidly in healthcare. This article shares some great examples of how it can improve patient care. 1. AI in Medical Diagnosis 2. AI in Drug Discovery 3. AI in Patient Experience 4. AI in Healthcare Data Management 5. AI in Robotic Surgery OtterSoft specializes in helping to develop AI products for healthcare companies, enhancing efficiency and improving the patient experience. #ai #healthcare #digitalhealth #telehealth

Super proud and excited to share our own multimodal foundational model, Harrison.rad.1, 👏 the world’s most advanced multimodal foundational model designed specifically for the medical domain. #harrisonrad1 #foundationalmodel #ai #genAI #ArtificialIntelligence #MedTech #Healthcare #Radiology

Hot off the press! As artificial intelligence rapidly advances in healthcare, rigorous evaluation standards are imperative to ensure AI systems are safe, effective, and worthy of the high-stakes decisions they'll influence. Improper validation of medical AI could lead to erroneous judgments, putting patients at risk. We must develop holistic evaluation frameworks centered on: 🔒 Safety & Ethics: Assessing performance on diverse, real-world data representing different populations. Testing edge cases and confirming reliability. 🤝 Collaboration: Partnering software engineers, data scientists, and medical practitioners to ground evaluations in clinical realities. ✅ Interpretability: Advancing transparency into AI model judgments and limitations through techniques like explainable AI. 🥼 Clinical Relevance: Selecting evaluation metrics tuned to the specific medical context and patient outcomes. 🔍 Oversight: Implementing clear processes for ongoing monitoring, updating, and responsible deployment. As AI capabilities grow in areas like disease diagnosis, treatment selection, and risk prediction, health leaders must uphold stringent standards. Peer-reviewed journals should set high bars for disclosing evaluation practices. What are your perspectives on validating AI for these high-stakes use cases? I'd value your insights. A practical guide to the implementation of AI in orthopaedic research, Part 6: How to evaluate the performance of AI research? Ayoosh Pareek, MDPriv.-Doz. Dr. med. Philipp Winkler, Bálint Zsidai, Eric Hamrin Senorski, Sebastian Prof. Kopf, M.D., Ph.D., Christophe Ley, PD Dr. med. univ. Elmar Herbst, PhD, Jacob Oeding, Alberto Grassi, Michael Prof Dr. med. Hirschmann, Volker Musahl, Kristian Samuelsson, Thomas Tischer, Robert Feldt https://lnkd.in/eD7UGywj