I’ve been playing with the CAMELYON dataset of WSIs and I’ve trained a neural network to predict with degree of certainty if a tile from the given virtual slide has metastasized cells. Combining all predictions over the entire slide with a threshold cutoff produces a mask depicting ROIs that the model thinks deserve attention:
I would like to have an integrated view of the virtual slide, model predicted annotations, be able to edit or add new annotations and labels/metadata so I can assess the apparent quality of the model and also share it in simple UI with others, instead of just sharing the TensorFlow model/weights with Jupyter notebooks.
Essentially, I’m looking for a software/suite that has at minimum the following:
- A WSI Viewer
- Annotation tools
- An API to allow programmatic access to image tiles, annotations, metadata, etc of the image plus the ability to add update them
- Has good documentation and community support
- User authentication and being Web-based tool is a plus
I’ve searched and come across the following packages that seem to fit my requirements but I’ve had issues with:
- Cytomine - this one is quite impressive as it boasts a lot of cool features however, I’ve found the documentation for it to be quite sparse and I do not wish to pay for consultation services given that this is not a commercial endeavor and moreover Cytomine is supposed to be an Open Source project. (I was able to install the docker containers but got stuck when trying to upload an image; couldn’t find a solution to this issue online, yet). The devs have created a wonderful behemoth of a Software that overwhelms me due to my lack of programming experience. If there was more community support for it I would love to have used it (a dedicated moderator could perhaps help jumpstart one)
- QuPath - I haven’t spent a lot of time with this package since it is not geared towards WSIs primarily and the API wasn’t very clear to me (I use Python and prefer if the API was language neutral like REST)
- ASAPAnnotation - quite straightforward but lacks documentation and an API that would allow me to add and run my model
- OMERO - still investigating it but at the moment unable to view a ~2GB image (something viewers like Openseadragon are able to do quite seamlessly - I’m probably missing something here or it may not be optimized for WSIs…?)
My hopes for creating this topic are to save time (for myself and others) in searching for the right Software and help move my project forward.
Thank you for your time.