Clalit Innovation’s Post

Extracting medical information (medications/symptoms) from free-text and unstructured patient-generated health data is important in healthcare AI, where LLMs play a huge role. I also use LLMs to extract biomarkers from pathology reports (and if they are mutant), and ICD codes or disease symptoms from RWE data and then feed them into an ML model. It seems NuExtract, a 3.8B fine-tuned model based on Phi-3, is a great option for that. It is claimed to be better than GPT-4o in the English zero-shot benchmark while being 500 times smaller. One thing is to remember, the model does not generate anything new; it only extracts structured information (copy-paste parts of the document). Take a look at the notebook to see how it works. I have used Ollma, but you can use the model from Hugging Face, which is a little slow during inference. I have also added an extra example, where NuExtract can be used to extract details from a CV. Notebook: https://lnkd.in/exQucP_Z #DataScientist #MachineLearning #LLM #AI #AIforHealth #RWE #EHR

Omi Group recently hosted a groundbreaking seminar on leading healthcare Generative AI (Gen-AI) in Korea! The event focused on revolutionizing the medical industry by maximizing diagnosis and treatment efficiency, a significant leap forward in healthcare innovation. Industry experts explored the transformative potential of Gen-AI, discussing how this technology can enhance patient outcomes in Korea. Key highlights included: - Practical applications of Gen-AI in diagnosis and treatment - Case studies demonstrating improved patient care and operational efficiency - Discussions on seamlessly integrating Gen-AI into existing healthcare frameworks Join us in driving innovation and improving healthcare services. Omi Group's dedication to leveraging advanced technologies like Gen-AI reflects our commitment to creating a healthier future for all. Explore more details in the full article: https://lnkd.in/gm23fDTY #HealthcareInnovation #GenerativeAI #AIinHealthcare #KoreanMedicalIndustry #OmiGroup #Ominext #OmiKorea #FutureOfHealthcare #TechInnovation

Part 5 AI Models: What Makes Them Work? 4️⃣ Multi-center datasets, like the CONFIRM registry ( J Cardiovasc Comput Tomogr 2018), are essential for generalizable models across populations. 🌍 5️⃣ Data should be split into training, validation, and testing sets for unbiased performance. 6️⃣ Beware of image compression and post-processing—models may struggle to generalize across centers. 7️⃣ Missing data isn’t random—use proper methods to fill the gaps. 📉 8️⃣ Lack of standardization in storing clinical images makes AI adoption a challenge. Let's push for change! 🏥 Have you implemented an AI model in your #cardiacimaging department? Even for simple tasks or measurements? Share your methods, success and challenges here! #ArtificialIntelligence #CardiacCT #AIinCardiology #MedicalImaging

With my new subscription to NEJM AI (https://ai.nejm.org/) came a surprise gift, a collection of articles from the NEJM, NEJM Catalyst Innovations in Care Delivery, and NEJM Evidence on artificial intelligence and machine learning in medicine. Here are a selected highlights: 1️⃣ While AI has made significant progress in clinical applications, establishing norms, addressing bias, and standardizing the evaluation of AI interventions remain key challenges. (Haug, C. J., & Drazen, J. M. 2023. Artificial Intelligence and Machine Learning in Clinical Medicine) 2️⃣ Radiology is at the forefront of AI application, with AI assisting in image analysis, triage, and diagnosis. However, issues of generalization and post-deployment monitoring need attention for broader clinical use. (Rajpurkar, P., & Lungren, M. P. 2023. The Current and Future State of AI Interpretation of Medical Images.) 3️⃣ AI plays a crucial role in monitoring and predicting infectious diseases, offering real-time data analysis that supports outbreak management and public health initiatives. (Brownstein, J. S., Rader, B., Astley, C. M., & Tian, H. 2023. Advances in Artificial Intelligence for Infectious-Disease Surveillance.) 4️⃣ GPT-4 chatbots hold promise for medical documentation and assisting with differential diagnoses but require careful oversight to ensure the accuracy and relevance of their outputs in clinical settings. (Lee, P., Bubeck, S., & Petro, J. 2023. Benefits, Limits, and Risks of GPT-4 as an AI Chatbot for Medicine.) 5️⃣ AI can facilitate teamwork in healthcare settings by improving workflows, such as advance care planning and reducing unplanned care escalations, thereby enhancing patient outcomes. (Li, R. C., Smith, et al. 2023. Using AI to Empower Collaborative Team Workflows: Two Implementations for Advance Care Planning and Care Escalation.) 6️⃣ AI has demonstrated potential in real-time cancer screening, assisting clinicians in distinguishing between benign and malignant colonic polyps, which can improve diagnostic accuracy. (Barua, I., et al. 2022. Real-Time Artificial Intelligence–Based Optical Diagnosis of Neoplastic Polyps during Colonoscopy) #AI #ML #medicine #healthcare #futureofhealthcare

A study came out a few days ago that looked at the diagnostic accuracy of AI in DP A rapidly evolving area that seeks to enhance diagnostics using whole slide images The results showed that AI solutions applied to whole slide images had a high diagnostic accuracy overall (sensitivity of 96.3%, specificity of 93.3%) 𝗦𝗼, 𝘄𝗵𝗮𝘁 𝗱𝗼𝗲𝘀 𝘁𝗵𝗶𝘀 𝗺𝗲𝗮𝗻? The high sensitivity and specificity suggest that AI has the potential to: •• Improve diagnostic accuracy and consistency compared to human pathologists •• Streamline the pathology workflow by automating certain diagnostic tasks •• To reduce variability and subjectivity in pathological assessments •• Provide decision support for pathologists (especially in complex cases) Despite the potential, more research is needed to validate AI’s utilization in DP before it can be integrated into clinical practice 𝗞𝗲𝘆 𝗮𝗿𝗲𝗮𝘀 𝗳𝗼𝗿 𝗮𝗱𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗶𝗻𝗰𝗹𝘂𝗱𝗲: 1) Broader evaluation across a range of diseases and sub-specialties to ensure reliability 2) Whether the AI model can consistently perform across different patient demographics and healthcare environments 3) Assessing how AI can be integrated into existing workflows and its impact on clinical decision-making 💡 As we advance, the focus must be on rigorous validation, ethical considerations, and foster a synergy between human expertise and AI. #AI #healthcare #diagnostics #digitalpathology [Link to Full Study: https://lnkd.in/eXjBQEkV]

EBM Health Consultants Posters presentations and Poster tour, May 6th, 2024! Meet us at Poster board #5046 and #6007 to discuss the need to perform diagnostic meta-analysis and the use of #singlearm trials with aggregate data in three level Bayesian Hierarchical #NMA model Visit Poster Tour Area B, Level 1, Poster board 68 between 15:30-16:15 to learn more about the use of large language model in expediting the Risk of bias assessment in clinical SLRs. #ISPORAnnual #EBMHealth #ISPOR #HEOR #metaanlysis #AI #LLM #ChatGPT

What if AI could catch diseases before they spread and save countless lives? Traditional AI methods often fall short, missing out on the deep knowledge that specialists have about how our bodies function on a cellular level. This inspired me to become a part of 𝗣𝗮𝘁𝗵-𝗥𝗔𝗚: 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲-𝗚𝘂𝗶𝗱𝗲𝗱 𝗞𝗲𝘆 𝗥𝗲𝗴𝗶𝗼𝗻 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 𝗳𝗼𝗿 𝗢𝗽𝗲𝗻-𝗲𝗻𝗱𝗲𝗱 𝗣𝗮𝘁𝗵𝗼𝗹𝗼𝗴𝘆 𝗩𝗶𝘀𝘂𝗮𝗹 𝗤𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗔𝗻𝘀𝘄𝗲𝗿𝗶𝗻𝗴, a project we proudly submitted to 𝗠𝗜𝗖𝗖𝗔𝗜(Medical image computing and computer-assisted intervention, a significant platform in healthcare technology) 2024. We didn’t just improve AI accuracy, we supercharged it. We boosted diagnostic precision from 38% to 47%, with an impressive 28% leap for certain key pathology images. These figures are more than just statistics each one represents faster, more accurate diagnoses means doctors can catch diseases earlier and start treatment sooner, leading to better outcomes for patients. This work is transforming diagnosis using AI to save lives with a real impact on people around the world. Ready to be part of something big? Follow along, share your thoughts, and let’s explore how we can push the boundaries of what AI can do in healthcare together. I’ll be sharing more of this exciting journey, and I’d love to hear your story too. #RAG #AI #ArtificialIntelligence #PathRAG #AIinHealth #MICCAI

Don't miss out on this insightful article in Calcalist כלכליסט, written by Eran Lerer from Shoni Health Ventures, about how AI is revolutionizing the field of digital health, and what will make Israel a global leader in the field.