Autoimmune diseases are an important field of medicine and sometimes it is necessary to identify cellular patterns to correctly diagnose an illness. This project aims to create a network that classifies specific antibody patterns for those kinds of diseases.
Initially some types of image processing techniques were applied, which in hand with a cellular segmentation package called Cellpose aimed to automatically identify and segment every cell in a given image, using a U-net variant. Those cells were pre-processed and given to a neural network to train with. Multi-label and clusering methods were also studied for this problem for a more correct understanding of the cells. With it, a path for inputting and classifying external images was made.
This repository contains:
-
The report explaining the work produced and its motivation.
-
The code produced for this study: The code is ordered here using out image treatment pipeline, where the results from the previous work is used in the next.
- Cellpose_nucleus_test
- Segmented_nucleus_saved
- Image_nucleus_study
- Cellpose_cyto_test
- Segmented_cyto_saved
- Image_cyto_study
- CNN_patterns
- CNN_HEP
- MultiLabel_CNN_HEP
- Interimage_cell_clustering
As well as :
- Folder Guides - Guides on how anyone can make their own system by reusing ours
- Folder Old notebooks - Notebooks that were exploratory and fundamental to the main notebooks above
-
The dicts generated from them: They represent the avarage diameter of a cell within each class of antinuclear antibody pattern, for just the nucleus, as well as with the cytoplasm.
- Avarage_Cyto.csv
- Avarage_Nucleus.csv
- Avarage_Pattern_Cyto.csv
- Avarage_Pattern_Nucleus.csv