In Nature today, our work on the first foundation model for whole-slide digital pathology. Key advance is it shows strong evidence of scaling benefit. State-of-the-art performance on 17/18 imaging tasks, with 12 showing very significant improvement. https://lnkd.in/gRzZgqba
Ecellent work Peter Lee and Team. I will be looking forward to read the paper in detail, esp GigaPath and LLaVA-Med code base seems really interesting and surely it will be a great example to see both working together.. Is the GigaPath code/model open source?
Awesome work! We featured it on #brainxcommunity too: https://brainxai.org/learn/
Congratulations, Peter Lee, on this groundbreaking achievement in digital pathology! 🌟 It's impressive to see state-of-the-art performance on 17 out of 18 imaging tasks, with 12 showing significant improvement. How do you foresee this scaling benefit impacting the future of medical diagnostics? 🔍 ✍️ As always, I hope this provides you value! 🔔 Follow me for more about #Informationsecurity & #Riskmanagement
Exciting work, Peter! Thank you for leading the charge for the industry!
Thanks for sharing Peter!
Peter Lee , congrats! Exciting work!
Thanks for sharing
Co-Founder, PathPresenter, Director of Dermatopathology and Digital Pathology, Summit Health
1moSpectacular work and thanks for sharing the code and weights. Although quite a few foundation models have recently been published for pathology images and some for images and text, it is still difficult to justify the costs as usecases are not very obvious. If we want to use them to build specific models for clinical use and help in arriving at a better and faster diagnosis, multiinstitutional studies are lacking for the same. Even if the model is 95 percent accurate, who takes responsibility for the remaining 5 percent. An experienced pathologist may be able to override the result, but a pathogist right out of training may end up with disastrous reults if they believe the AI output. In the pharma space, the use of foundation models related to pathology images has been touted as helping finetune smaller models for predicting underlying molecular alterations. However the prediction from current models is similar to what was achieved using CNN. How does one then justify the costs to switch from CNN to FM. Work still needs to be done till we find financilally viable and true usecases in healthcare. Maybe more can be acheived if all this great work by organizations be done in collobaration rather than as academic exercises in silos,