Look back
- Prepare write
- Literature Review(14) and Project Plan
- Ethical Assessment Form (if needed)
- Risk Assessment (if needed)
- Cellpose 1/2/3/SAM
- AIM-CICs(2022) an automatic identification method for cell-in-cell structures
- detect: Faster-R-CNN + ResNet-50 & classify: ResNet-101
Look forward
- Continue finding and reading literature
- Unet method
- CellViT
- traditional method paper
- Finish formative assignment
21.5.2025
%% – %%
Look back
- AIM-CICs
- Detection + Classification (not segmentation)
- with code. no model weights (already email)
- Try to use Cellpose-SAM
- Learn to use ImageJ
- Instanseg August 29, 2024
Look forward
- Image label
- Train and test a model based on Cellpose-SAM to ”panoptic segmentation”
- Try other segment model
27.05.2025
Look back
- Labeling
- Tried YOLOv12
- not very good
- 90 images (70% train 20% cal 10% test)
- Data Augmentation
-
Label data for cellpose tarining (Not yet finished)
Look forward
- Try more method (CellViT)
- Data Augmentation
04.06.2025
keep on searching papers
Look back
- label data for cellpose-sam
- train models in cellpose-sam
- some case good
- some case bad
- The cells can be found, but they are closer to a circle, closer to the center
- Read paper
- MedSAM
- MedSAM2
- MedSAM Adapter
Look forward
- Try training with different parameters
- Try more method (CellViT)
- Understand paper
11/6
—
Look back
- cellpose-sam trained
- average precision at iou threshold 0.5 = 0.704
- Focus on nucleus
- Hover-Net H&E color different with Fluorescence staining
- Cell-Vit some problems with the ncc environment configuration
Look forward
- CellViT/Deepcell
- Detection of nucleus
- Detection of cell membranes
18/06
—
Look back
- CellViT
- environment done but image form need .wsi(.tiff etc.)
- The previously trained models were deployed locally and online
Look forward
- Wait for more data from Mary
- Separately segement the cell membrane and nucleus, and then judge based on the morphology.