🚀 Big news! Slideflow 3.0 is here! 🎉 We've revamped our licensing (now Apache-2.0) to make Slideflow more open & accessible. We've added 3 foundation models (Virchow, UNI, GigaPath), upgraded MIL, & much more. Check it out: https://t.co/IlxP0jEmLs #AI #Pathology #OpenSource

🎉Congratulations to the 2024 #CZBiohubCHI Investigators! The cohort of 48 of Chicago’s most creative and accomplished scientists and engineers will use multidisciplinary approaches to understand inflammation and the dysregulation of immune cells.

@lab_pearson @fredhow @sidd @SaraKochanny @andrewsris1 @GarassinoMarina For inquiries and discussions about collaborative opportunities, please reach out to james@slideflow.dev!

Slideflow has been a labor of love, born out of research at University of Chicago with @lab_pearson, @fredhow, @sidd, @SaraKochanny, and so many more. Special thanks also to the brilliant @andrewsris1, as well as @lab_pearson & @GarassinoMarina, who have been wonderful mentors!

We'd love to hear from you - please reach out if you have any questions, ideas, or are interested in collaborating with us to support the next generation of computational researchers! GitHub: https://t.co/Yme4VdhIAy Docs: https://t.co/CbFdHC2Ik2 Paper:

Our accessible user interface, Slideflow Studio, supports live model deployment on whole-slide images and provides research tools for interrogating how models behave. https://t.co/lvQBrTviI6 (4/5)

Slideflow supports a wide range of digi path tasks: - Predictive models trained to categorical, continuous, or time-series outcomes - Uncertainty quantification - Tissue & cell segmentation - Self-supervised learning (SSL) - Generative adversarial networks (GANs) ... (3/5)

Slideflow is designed for building, validating, and deploying reliable and explainable AI biomarkers. With an emphasis on flexibility and ease of use, our aim is to democratize access to these highly complex model building paradigms and facilitate innovation. (2/5)

I'm thrilled to share our new manuscript presenting Slideflow: a state-of-the-art deep learning toolkit for digital pathology. Train your own DINOv2 foundation model, transformer MIL biomarkers, and explainable GANs. Easy integration with external models.

Super exciting research from @andrewsris1, presented at ASH this year!

Huge congrats, @lab_pearson! Well deserved recognition

Thanks @SaraKochanny! And special thanks to the entire @lab_pearson as well, without whom Slideflow would not exist!

🔬 Slideflow 2.2 is now available! Our accessible deep learning toolkit brings powerful AI tools to your fingertips. Now with multi-magnification MIL, DINOv2 support, and new pan-cancer pretrained feature extractors. See our 📖 docs for more info! https://t.co/CbFdHC2Ik2

🔬Slideflow 2.1 is now available! Easy AI for #digitalpathology, now with interactive MIL attention heatmaps, more robust feature extraction tools, and optimized slide QC. 📖Docs: https://t.co/rJEafYUYGT 📝Release notes: https://t.co/v8yAr0ToTb

Thank you! It’s an honor to have been selected, and I greatly look forward to continuing our exciting research in AI for oncology alongside @lab_pearson, @marinagarassino and @fredhow!

Honored to participate in this secondary multiomic analysis of the PEOPLE trial, looking for predictors of immunotherapy response in advanced NSCLC, with @PrelajArsela and @marinagarassino. Looking forward to carrying this work forward as we study IO response through I3LUNG!

This work wouldn't be possible without a broad and diverse team - Rachelle Wolk, @hieromnimon @fredhow @andrewsris1 and so many more! It's been an honor working with @lab_pearson and @nicolecipriani at @UCCancerCenter. Lots of exciting work ahead using GANs for #digitalpathology!

Synthetic histology can also be used to for education. We asked pathology residents to predict BRAF-like (classic PTC) vs RAS-like (NIFTP) for thyroid cancers at U of C (a challenging task!). Accuracy significantly improved after 1-hr of teaching using only synthetic histology.

Using synthetic histology, we discern nuanced morphologic differences between ER(+) and ER(-) breast cancer, HPV(+) and HPV(-) head & neck cancer, and BRAF_V600E-like and RAS-like thyroid cancers. The same method coulud be used to illustrate molecular states in any cancer. (3/5)

Classifiers make predictions from images. Generative adversarial networks (GANs) create realistic, synthetic histology images. Used together, we can open "black box" classifiers, facilitating model transparency & discovery of how molecular states manifest morphologically. (2/5)