This work include cell instance segmentation based on Cellpose@https://github.com/MouseLand/cellpose, using CNN to classification cell's phenotype during mitosis period.
This project mainly consist by:
- Segmentation part ( using cellpose model )
- Classification part (including classification model's construction & training )
- Processing part (including a complete inference and plot )
We using cellpose to segment cell image. Note the code in the cellpose folder is not as same as original cellpose project. The model weight is directly download from original cellpose project.
For the classification part, we implement and train a CNN to classification the phenotype of cell. The model archtecture and training procedure can be found in the clsmodel.py , training data can be found in the ==cell cognition== website.
In this project, we use H2b_aTub_MD20x_exp911 dataset, H2B annotation, approx 1000 pictures. We treat them as follows:
- iterate each sample, find them in the original experiment image.
- crop 64*64 picture on each sample's center
- apply Contrast Limited Adaptive Histogram Equalization
- apply flip augmentation if needed
You can find perpared training data at ./dataset/mc64 folder.
complete processing and plot can be found in test.py