I am working in the area of Digital and Computational Pathology which deals with whole-slide imaging including creation of patches, annotation, semantic segmentation, classification, weakly-supervised learning, active learning, vision transformers etc.
There are several open-source tools available like Viewers (QuPath, Digital Slide Archive etc.) and analytics frameworks like the Medical Open Network for AI (MONAI) framework.
I would like to use KNIME as the core platform for whole-slide image analysis using some of the tools mentioned above. But tools like MONAI are based on the PyTorch deep learning framework, whereas KNIME currently supports Tensorflow/Keras for deep learning.
While reading of large whole-slide images is possible using the OMERO extension in KNIME, bio-image processing in KNIME could be further strengthened using the PyTorch deep learning framework and MONAI.
Hence, it would be nice to have an extension for PyTorch based deep learning, in addition to the Keras/Tensorflow based deep learning extension already available.
Does anyone have a similar wish-list or a concrete plan to integrate PyTorch for deep learning?