New paper out in npj Health Systems! An ensemble of local, open-source LLMs can automatically monitor clinical AI performance in real time — no radiologist review needed. The foundation of our RADAR system. 🧵
https://t.co/bZ1TIo5MRw
#RadiologyAI#LLM#HealthInformatics
⚠️ AI Hallucinations: What Every Developer Needs to Know 💡
AI hallucinations aren't just glitches—they can lead to significant risks, from downtime costs to legal issues. For #AI developers working with #LLMs, it's essential to understand how to detect and prevent these errors to build trustworthy models. Here’s what you need to know. 🧵 [1/3]
Read the perspectives of experts from @MICCAI_Society and @RSNA on the clinical, cultural, computational, and regulatory considerations to adopt #AI technology successfully in radiology https://t.co/CsPEVa2O1R #AIME2024
Coming soon: #SIIM chair @asset25 will share an exclusive, behind-the-scenes look at SIIM's priorities at the #SIIM24 Membership Meeting! Join us to celebrate SIIM Awardees & meet SIIM Leadership!
Jun 28 | National Harbor, MD
https://t.co/eBDetJU2ku
All the CheXpert Plus links in one place:
--Dataset: https://t.co/9ZkvmBTmbg
--Arxiv paper: https://t.co/ggwx6QOS74
--De-ID algorithm paper: https://t.co/DeQvoo9tsw
--Hugging Face De-ID algorithm: https://t.co/28n6hq2X6s
@StanfordAIMI
Contrastive learning and the study of algorithmic fairness require more and different CheXpert data. To meet this need, we are releasing CheXpert Plus, including radiology reports, demographic data, DICOM images, pathology labels, & RadGraph extractions. https://t.co/qtRn3iZB0a
Five years ago, thanks to the leadership of @mattlungrenMD, @stanfordAIMI released the CheXpert images: 223K JPG CXRs with labels for 14 conditions. CheXpert has been cited >6000 times, mostly related to development of supervised learning methods. Much has changed since then.🧵
Join us today!
Our May seminar is happening today at 2pm CT and featuring research on "The MIDRC Diversity Calculator: A Dynamic Tool for Measuring and Monitoring the Representativeness of Biomedical Datasets."
Registration here: https://t.co/Ecgyuii3Am
Many-Shot In-Context Learning in Multimodal Foundation Models
🤔 They benchmarked GPT-4o and Gemini 1.5 Pro using 10 datasets and found that many-shot ICL results in significant improvements compared to few-shot ICL across all datasets.
https://t.co/ERSX75DNUi