In this repository, we've implemented the code resulting in the submitted paper "Deep-Fuzz: A synergistic integration of deep learning and fuzzy water flows for fine-grained nuclei segmentation in digital pathology." in Plos One, by Nirmal Das, Satadal Saha, Mita Nasipuri, Subhadip Basu, Tapabrata Chakraborti. The Deep-Fuzz code repository conssist of two different module: 1) a U-Net based deep learning module and 2) Qt C++ based fuzz module.
The work is focused on segmentation of overlapping nuclei from microscopic images. In the first phase we used U-net based deep learning module with ASPP block in the bottleneck for initial coarse segmentation. In the second phase, we used a novel fuzzy waterflow algorithm for instance level segmentation of the overlapping nuclei.
- For the deep module, we have adopted the code at: https://github.com/nauyan/Segmentation with hyper parameters tuning. The installation gudie can also be found at the given site. The installation gudie is given at the site and the required libraries can be found here.\
- To run the fuzzy waterflow module you need to install QT software. The fuzzy waterflow code files are uploaded here as flow.rar. The code is written in C++.
We used four public datasets: NuCLS, MoNuSeg, TNBC, and S-BSST265.