If you want to learn about #bioimageanalysis I've written a free & open textbook that tries to help:. Thanks to the wonder of @ExecutableBooks & other modern magic it's not quite like a normal book. (thread). @OpenEdEdinburgh @NEUBIAS_COST @BioimagingNA

RT @EuroBioImaging: Join us for the #VirtualPub with @petebankhead, University of Edinburgh, for a talk entitled "@QuPath for bioimage anal….

RT @QuPath: Find out about the latest QuPath developments as part of the @GloBIAS_ series 'Current and future plans for essential BIA tools….

RT @NerdyCammy: Answering the call of @haesleinhuepf, @tischitischer,@petebankhead,@cmci_,@CiminiLab . My @FijiSc training notes are now FA….

RT @LJIMicroCore: 📢Announcing a QuPath Workshop 📢 February 24-25, 2025, in San Diego, CA, at the La Jolla Institute for Immunology. This y….

RT @frank_pado: 🚨 New segmentation model alert! 🚨🥳 . You can finally use InstanSeg in our #BioImageAnalysis software #CellACDC! . InstanSeg….

RT @Path_Matt: We remixed holiday classics with an AI twist—check it out! .All I Want for Christmas is Digital Pathology.Frosty the Cryosta….

Find out more about InstanSeg from Thibaut himself at his Virtual I2K tutorial next week! .Register at

RT @QuPath: Join us at the Virtual I2K meeting next week!.Lots of image analysis tutorials - including 3 about QuPath & 1 about InstanSeg….

Second QuPath v0.6.0 release candidate now available for testing at This speeds up InstanSeg for complex annotations *and* adds much improved support to read & write OME-Zarr - @openmicroscopy.

RT @haesleinhuepf: "Despite repeated insistence from experts that sharing AI training data, code, and model weights [. ] are necessary bes….

❗️Important info hidden in the long thread❗️.Download the new QuPath v0.6.0 *release candidate* for the easiest way to try this out 💻🔬🤖.

RT @SaraMcArdle1: I literally squeaked when I tried InstanSeg myself. I'm worried I frightened @petebankhead and Thibault Gold with my enth….

. and now continued here:.

To summarize: we think InstanSeg could be pretty useful - for us & hopefully for others. Learn more about the method at And see how it extends to multiplexed images & whole cell segmentation at.

This bundles up QuPath & the InstanSeg extension, so it's easy* to get started. Download it at *-Sorry non-Apple-silicon-users, GPU support can still be a pain to set up. Our attempt to help is at

The QuPath extension requires QuPath v0.6.0… which won’t be released until October (possibly seconds before the workshops start at ). So today we've made an *early release candidate* available for anyone who wants to try it out in advance.

But, as always, it's better to try for yourself & see which methods work best for you. So how can you get all this?. Find the Python code & DeepImageJ-friendly models at

Comparisons are tricky, because we don't know of another method that's as versatile for multiplexed images. We explore a few angles in the preprint. Our results suggest InstanSeg is more accurate & much faster than Mesmer, while lifting the restriction on 2 input channels.

It'll also let you choose channels if you really want to (e.g. for speed, or because you know they aren't informative). Or even use color deconvolution, if you want to try applying a fluorescence-trained model to a brightfield image.

We share a model trained on the Aleynick et al. dataset using a fixed resolution of 0.5µm/px. If the pixel size info is found in your image, the QuPath extension will automatically rescale everything internally to run at the expected resolution - and rescale the output.