EdgeSSFewShotMicroscopy

Repository for Edge-Based Self-Supervision for Semi-Supervised Few-Shot Microscopy Image Cell Segmentation

Abstract

Deep neural networks currently deliver promising results for microscopy image cell segmentation, but they require large-scale labelled databases, which is a costly and time-consuming process. In this work, we relax the labelling requirement by combining self-supervised with semi-supervised learning. We propose the prediction of edge-based maps for self-supervising the training of the unlabelled images, which is combined with the supervised training of a small number of labelled images for learning the segmentation task. In our experiments, we evaluate on a few-shot microscopy image cell segmentation benchmark and show that only a small number of annotated images, e.g. 10% of the original training set, is enough for our approach to reach similar performance as with the fully annotated databases on 1- to 10-shots. Our code and trained models is made publicly available.

##Requirements

Step 1) Create environment

Install environment from env.yml using conda

conda env create -f env.yml

Step 2) Download Datasets

Place each dataset inside 'Datasets/Raw/DATASETNAME/' Note: Names of datasets are B5, B39, TNBC, EM, ssTEM

Step 3) Run preprocessing pipeline

Run the code inside run_preprocessing.py to pre-process source and target data.

Code

Train with Edges

Example

python Learning_main.py  --target 'TNBC' --affine 0 --lr 0.001 --loss 'weightedbce' --name 'miccai_jointOpt_cannyedge_10percentlabelled_100percentunlabelled'

Fine-tune model

Example

python Evaluation_main.py  --eval-selections  1 2 3 4 5 6 7 8 9 10   --lr 0.001 --finetune-epochs 20 --finetune-lr 0.0001 --finetune-loss 'weightedbce' --architect 'FCRN' --switchaffine 1 --num-shots 1 3 5 7 10   --statedictepoch 50 --target 'TNBC'   --pretrained-name $YOUR_PRETRAINED_MODEL_NAME --finetune-name $FINETUNED_MODEL_NAME$

Our pretrained models

Will be uploaded soon!

Cite

Please use the following citation:

@inproceedings{dawoud2022edge,
  title={Edge-Based Self-supervision for Semi-supervised Few-Shot Microscopy Image Cell Segmentation},
  author={Dawoud, Youssef and Ernst, Katharina and Carneiro, Gustavo and Belagiannis, Vasileios},
  booktitle={International Workshop on Medical Optical Imaging and Virtual Microscopy Image Analysis},
  pages={22--31},
  year={2022},
  organization={Springer}
}