Sapio Sciences’ Post

Chest X-rays, the most frequent clinical imaging test, can now be tackled by AI thanks to a new open-source dataset, LLM, and benchmark 🫶 😮 Chest X-rays (CXRs) are the most frequently performed imaging test in clinical practice. Why developing LLMs to interpret CXRs is hard: 1️⃣ Limited availability of large-scale vision-language datasets in the medical image domain. 2️⃣ Lack of vision and language encoders that can capture the complexities of medical data. 3️⃣ Absence of evaluation frameworks for benchmarking the abilities of LLMs on CXR interpretation. What this paper introduces: 1️⃣ CheXinstruct - a large-scale instruction-tuning dataset curated from 28 publicly-available datasets.  2️⃣ CheXagent - an instruction-tuned LLM capable of analyzing and summarizing CXRs. 3️⃣ CheXbench - a novel benchmark designed to systematically evaluate LLMs across 8 clinically-relevant CXR interpretation tasks. —————————————————— Want to stay at the forefront of Generative AI developments? Follow NLPlanet for daily insights into the most relevant news, guides, and research! 🚀

The New England Journal of Medicine has launched a new journal, NEJM AI, geared toward ensuring the responsible development of AI in healthcare. Of particular note is the need for new interdisciplinary teams to develop, evaluate and bring to market AI approaches in healthcare. As the NEJM AI editorial puts it: "In our eyes, the most impactful articles will blossom from the fertile ground of multidisciplinary teams, reflecting the vibrance at the intersection of computer science, clinician–patient dynamics, and biomedical research." We're seeing many collaborations of this sort across Carnegie Mellon University, University of Pittsburgh, UPMC, and Allegheny Health Network, with more to come! https://ai.nejm.org/

High quality data input for AI, applied for drug discovery! See you next Wednesday in Basel! #ai #drugdiscovery #ml #pharma #biodata #knowledgegraph #opentargets #bioinformatics

⏰ What could 902 extra hours do for your AI development timeline? Our longtime customer Activ Surgical just published a paper in Nature Portfolio that demonstrates the quality and time-saving benefits of annotating medical data on the Centaur Labs gamified crowdsourcing platform. The paper reports the first complete and adaptable methodology (powered by Centaur Labs) to obtain highly accurate segmentations of surgical tissues using non-expert crowdsourcing. 🎯 💸 The old labeling process: pay a handful of doctors a lot of money to deliver results on a long timeline. ✨ The Centaur Labs labeling process: launch your labeling tasks as a global contest to an engaged crowd of 50k+ skilled labelers and get results in as little as 24 hours. 🔗 Read the Nature article here: https://hubs.li/Q02vM-2D0 🔗 Get in touch to start your project: https://hubs.li/Q02rZ3DF0 Congratulations to the authors Garrett Skinner, Tina Chen, Gabriel Jentis, Yao Liu, Chris McCulloh, MD, Alan Harzman, Emily Huang, Matthew Kalady & Peter C.W. Kim, MD, CM, PhD.

Deep Learning and 7T MR, more than just better imaging... We are pleased to announce the start of a project funded by Bundesministerium für Bildung und Forschung to explore the combination of new deep learning methods and high-resolution anatomical and metabolic imaging at 7T. With a total budget of 2 million Euros, Integrated Deep Learning @ 7T (IDL@7T) aims to enable a strategic partnership between the Universitätsklinikum Erlangen and Siemens Healthineers in this highly innovative and competitive field. As part of the industry in clinic platform FastTrack MedTech, IDL@7T will run for three years, with the goal to improve diagnostic sensitivity and specificity in the future. #deeplearning #mri #7Tesla

🤩 Activ Surgical AI + Centaur Labs labeling = "Herein, we report the first complete and adaptable methodology to obtain highly accurate segmentations of surgical tissues using non-expert crowdsourcing."

A new paper published in Nature by one of Centaur's customers says we saved them 900+ hours of surgery video annotation. 😮 The way things used to be: "Crowdsourced annotations of surgical video ... have historically relied on unsophisticated crowdsourcing methodologies and have been limited to annotations of simple rigid surgical instruments and other non-tissue structures." 🚫 They explain how Centaur is different: "Here we describe an application of gamified, continuous-performance-monitored crowdsourcing to obtain annotated training data of surgical tissues used to train a soft tissue segmentation AI model." 🤓 Worth a read!

🎉 We're thrilled to announce the publication of our latest research paper, "Detection of Covid-19 and Pneumonia in Chest X-ray Images Using Deep Learning Techniques”. with contribution Tasneem Kandil which has been published in IEEE Xplore. The fight against COVID-19 requires rapid and accurate diagnostics, and our study enhances this capability by utilizing deep learning to distinguish between COVID-19 and pneumonia from chest X-ray images. This distinction is notoriously difficult due to the overlapping visual features of the two diseases in X-rays. 🔍 Key Highlights: Objective: Improve the accuracy of COVID-19 and pneumonia detection from chest X-ray images. Methodology: Employing the VGG-16 model, with and without image augmentation and enhancement techniques, on a dataset of 5970 X-ray images. Results: The enhanced VGG-16 model achieved an impressive accuracy of 92.72%, demonstrating significant improvements over traditional diagnostic methods. This breakthrough demonstrates the potential of deep learning to aid in the rapid and effective diagnosis of respiratory illnesses, which is critical in the ongoing battle against COVID-19 and other similar infections. 🔗 Read the full paper here: https://lnkd.in/d8_FPN-r Presented at the 6th International Conference on Computing and Informatics, our findings are a step forward in medical imaging and diagnostics, proving that advanced technologies can significantly impact public health. A big thank to all our colleagues and collaborators involved in this research. #COVID19Research #DeepLearning #MedicalImaging #HealthTech #XrayImaging #Pneumonia #ResearchPublication

“The next frontier [of AI in Healthcare] is how do we harness this power to drive new scientific and engineering advances that can help us address some of the most challenging problems in clinical research. For example, how do we connect different modalities to come up with a more holistic view of the health state or the disease state?” IBM’s Jianying Hu, PhD, discusses with Thomas Fuchs, DrSc., the future of machine learning and medicine in a special one-on-one discussion at The New Wave of AI in Healthcare symposium. Watch the full interview: https://lnkd.in/eKdfEZ8k