Technical Leader | Radically Curious | Innovating Virtual Care with Computer Vision | Epic Certified | Clinical Workflow Nerd | Author | Happiest On, In or By the Water 🌊 🚣♀️ 🏄♀️
A Year of Journal Articles (Day 10/365): Xu, Shawn, et al. "ELIXR: Towards a general purpose X-ray artificial intelligence system through alignment of large language models and radiology vision encoders." arXiv preprint arXiv:2308.01317 (2023) https://lnkd.in/gACK66GJ Summary: - Artificial intelligence systems for medical imaging have traditionally focused on very specific tasks and have performed inconsistently when leveraged to new problems. - The authors proposed a new approach that combines large language models and vision encoders to train multimodal models using routinely collected medical images and their associated text reports. This approach can be used to perform a diverse range of tasks, including zero-shot and data-efficient classification, semantic search, visual question answering, and radiology report quality assurance. -The authors evaluated their approach on several public and private datasets and showed that it achieved very strong performance on zero-shot chest X-ray (CXR) classification (AUC of 0.850), data-efficient CXR classification (AUCs of 0.893 and 0.898), and semantic search. It also showed promise on CXR vision-language tasks, demonstrating overall accuracies of 58.7% and 62.5% on VQA and report quality assurance tasks, respectively. - Compared to existing data-efficient methods, including supervised contrastive learning (SupCon), ELIXR required two orders of magnitude less data to reach similar performance. -Conclusion: LLM-aligned multimodal models can unlock the value of chest X-rays paired with radiology reports to solve a variety of previously challenging tasks. *I recommend reading this paper, while at first the 54 pages might throw you off its only about 28 before citations and there are a lot of good tables and pictures :) #AYearOfJournalArticlesBP
Value-Based Care Operator and Advisor for CEOs, CMOs, & CMIOs | I design scalable population health strategies incorporating people, process and technology to save lives, to expand access, & to reduce the cost of care.
10moBrittany P. another fabulous technical 📫