Akshay Chaudhari
@Dr_ASChaudhari
Followers
2K
Following
2K
Media
29
Statuses
491
Radiology AI | Faculty at Stanford | Co-Founder Cognita
Stanford, CA
Joined March 2019
A fine-tuned, open-source LLM (Mistral-7B) extracted clinical history from imaging orders, showing strong agreement with radiologists & rivaling GPT-4 Turbo. Dive into the findings: @magdapasc @Dr_ASChaudhari @stanfordradiology @stanfordaide #AI #Radiology
https://t.co/sUWySnaBEI
0
2
13
Does everything seem like a foundation model these days? We tried to add some rigor and definitions to help with the process! Kudos to @magdapasc for leading this work, which is now published in @radiology_rsna.
🧵 What if AI could learn from millions of unlabeled radiology images and reports—and then flexibly adapt to new clinical tasks? In a new comprehensive review in @radiology_rsna, we dive into how foundation models (FMs) are set to revolutionize radiology! @AIMI_Stanford (1/6) 👇
2
6
38
The most powerful agent, interpreting X-rays is opensource🤗
1/ Updates on our improved open-source CheXagent with new transparent benchmarks! We ran a new reader study mimicking real workflows: Radiology residents drafted reports that attendings reviewed/edited. Results from 8 rads show major efficiency gains. Key findings: 👇
1
2
7
8/ Paper, models, code can be found here:
stanford-aimi.github.io
CheXagent: Towards a Foundation Model for Chest X-Ray Interpretation
0
0
2
7/ This work was a true collaboration: Co-led by @zhjohnchan and @mayavarma23 Mentored by myself, @IAMJBDEL, and @curtlanglotz Grateful to @StanfordRadiology & @StanfordAIMI for their support. Excited for the future of AI in radiology!
1
0
0
6/ CheXagent provides a powerful foundation for tasks where priors are hard to learn from end-to-end training (e.g., captioning). We hope it inspires researchers to adapt it for new medical imaging applications.
1
0
0
5/ Models: 1. CheXagent, 2. RadPhi-2, 3. 8 CLIP/SigLIP CXR models.
1
0
0
4/ Supporting the research community, we are open-sourcing: Code for: 1. CheXinstruct curation 2. Model inference 3. Transparent benchmarking 4. Reader study interface
1
0
0
3/ We showcased CheXagent in action with a live demo at the Stanford Radiology Retreat. Radiologists selected cases and tested the model in real-time. This hands-on feedback is helping us refine how CheXagent integrates into clinical practice.
1
0
0
2/ - 36% time savings for residents using CheXagent-drafted reports. No sig. time difference for attendings editing resident vs. CheXagent drafts. Writing efficiency improved for 81% of residents and 61% of attendings. CheXagent may help improve efficiency of radiology workflows!
1
0
0
1/ Updates on our improved open-source CheXagent with new transparent benchmarks! We ran a new reader study mimicking real workflows: Radiology residents drafted reports that attendings reviewed/edited. Results from 8 rads show major efficiency gains. Key findings: 👇
2
7
27
Inspired by a few of my conversations with @Dr_ASChaudhari I decided to spend my winter break redesigning the stock Health app with an AI focused approach starting with clinical records. The goal here was to help patients get a deeper understanding of their own records
4
6
64
If you are at #NeurIPS2024, don't miss a chance to chat with @mayavarma23 on how to mitigate spurious correlations found in VLMs!
(1/4) Excited to share RaVL, which is appearing this week at #NeurIPS2024! RaVL discovers and mitigates spurious correlations in fine-tuned vision-language models (VLMs). 📄 Paper: https://t.co/5Xs2wKZ1Er 💻 GitHub: https://t.co/P2QbyreudD
1
1
15
So proud of the release of the GREEN metric. See what you can do when you merge medical AI research and open-sourceness. 26,000 downloads on 🤗 Hugging Face and counting. 🟥 EMNLP proceedings: https://t.co/GXohz9FGe0 🤗 Dataset: https://t.co/zTJtQiJnmV 🤗 Models:
3
10
30
We are assembling a lean team of engineers and researchers. If you're interested in making large-scale clinical impact on healthcare with AI, we would love to hear from you! Let us know here:
0
0
16
We are excited to announce the launch of our company - Cognita! We are working towards building the future of radiology through multi-modal AI systems with a great group of founders @loublanks , @zhjohnchan, and I, and advisors Ajit Singh, Chris Re, and @curtlanglotz 1/2
8
10
111
New @EurSpineJournal study introducing novel method for 3D analysis of lumbar paraspinal intramuscular fat. Method may deliver key insights into the interplay between #muscle #health and #lowbackpain. https://t.co/XPRZIsoaUG
@StanfordPain @PolyNeuro
0
12
27
1) Preference Fine-Tuning for Factuality in Chest X-Ray Interpretation Models Without Human Feedback presents a practical recipe for improving and scaling expert human quality report generation in high-stakes domains: using LLM-as-Judge for preference tuning in radiology report
1
2
8
Great work led by Dennis Hein with collaborators @zhjohnchan, @SophieOstmeier, Justin Xu, @mayavarma23, @edreisMD, Arne Michalson, @cxbln, Hyun Joo Shin, and @curtlanglotz. Work done at @StanfordRad @StanfordDBDS, @StanfordAIMI. 3/3
0
0
12
We explore different post-training approaches for VLM alignment, discovering interesting success and failure modes for both the underlying models and metrics. Full paper: Paper: https://t.co/WB6cuALTHy 2/3
arxiv.org
Radiologists play a crucial role in translating medical images into actionable reports. However, the field faces staffing shortages and increasing workloads. While automated approaches using...
1
0
8