We must maintain our momentum in order to attain all of our goals
Currently, we are still preparing the final bladder dataset with annotations, so weakly supervised learning methods have been developed without annotations to overcome the lack of a thoroughly annotated dataset in the meanwhile. Recently, I have been working on a grading algorithm for non-muscle invasive bladder cancer, developing the concept of nesting for weakly supervised learning methods and combining it with attention mechanisms for interpretability, motivated by the application of finding a specific diagnostic region in a sizable whole slide image. A novel machine learning architecture for weakly supervised learning methods was implemented and tested on classical and histological image datasets, and currently, a manuscript is under review. We are hopeful that the proposed architecture would be helpful in future applications related to WSI, and we could obtain future publications out of it.
We must maintain our momentum in order to attain all of our goals. I cannot stress enough the importance of developing research when you come up with a brilliant idea. If you have it, don’t let it go!
An attention-scored WSI for grading
Saul Fuster Navarro – ESR5.