Fig. 1
From: Deep learning for histopathological segmentation of smooth muscle in the urinary bladder
Summarized methodology. The annotated H & E-stained images are pre-processed and split into patches of defined size. These patches are extracted in two ways to create two datasets: ①patch-based: each patch is either fully MP or fully non-MP and ② pixel-based: each patch includes respective ground truth mask patch where white pixel corresponds MP and black pixel corresponds non-MP. Selected CNN models are trained on the patch-based dataset and deep learning models are trained on the pixel-based dataset. The trained models for patch-based approach and pixel-based approach will be used to semantically segment any given H&E-stained image into MP and non-MP regions