Hi everybody, thanks for opening the channel. I’m Caro and i recently started my PhD at QBiC in Tübingen and together with @Luis Kuhn, we work on different imaging-related projects there. One specific project is to build a nextflow/nf-core pipeline for the analysis of light sheet microscopy images. This is a collaboration with the group of Jason Stein at the University of North Carolina. They do light sheet microscopy of the mouse cortex and their data analysis is based on a toolbox called NuMorph (https://www.sciencedirect.com/science/article/pii/S2211124721012626?via%3Dihub) which is implemented in Matlab. The pipeline should replace the current Matlab pipeline and of course, simplify and improve their analysis. We didn’t propose the pipeline yet in the new pipelines channel, i thought maybe it would make sense to reach out here first since you @Konrad Rokicki already working on a light sheet image reconstruction pipeline. For me, it would be interesting to see the potential overlap of the nf-lightsheetrecon and our planned pipeline or what is already there that we could also use. I would be glad for any help or advice on that, to point me in the right direction, I’m very new to imaging and pipeline dev.

👋 @Carolin Schwitalla and @Luis Kuhn good to see you here ☺️ Happy to see more people join the microscopy discussion!

Pleased to meet you @Carolin Schwitalla and that looks like a cool project. Looking at the NuMorph toolset, there is indeed some conceptual overlap with the image processing tools we are using from the Saalfeld Lab and others (stitching, flatfield estimation, registration, etc.) You can read more about those tools here. If you see something there that you might be interested in reusing and/or extending, then I'd be happy to put you in touch with the tool authors. One thing I can recommend, especially if you are starting from scratch, is using next generation file formats (e.g. OME-NGFF) and a distributed processing framework like Spark or Dask. We've developed ways to run these frameworks inside Nextflow (and I'm in the process of adding these as nf-core modules). The "next gen" approaches make it possible to scale your processing up to much larger data sizes. The mouse brain images in the paper you cited are ~100 GB per channel. That's already at the edge of single node processing and usually the data only gets bigger over time. I think there are many touch points across our efforts, so it would be great to keep this conversation going!

Thank you for all the helpful information and i would be glad to stay in touch on this topic, maybe we could collaborate on some modules that would fit into both pipelines.

Dear @Konrad Rokicki thanks a lot for the useful pointers! specially on the image processing tools, it seems that there is indeed overlap with NuMorph. We also have OME-NGFF on our radar and we’ll make sure to use it (it seems like a necessity when dealing with these large data types). Regarding the distributed processing frameworks (spark, dask), it seems we also need to move in that direction as you mentioned, however we still need to do our homework to learn how to use them, I am sure on this topic we can learn a lot from your work. I also agree we should keep the conversation going.

Hi all. @Luis Kuhn, sorry for the slow response. Slack overload. Definitely looking forward to hearing how things work for all of you. Feel free to

@

-mention liberally 🙂

no worries, thanks a lot!