Adversarial Training Collaborating Hybrid Convolution-Transformer Network for Automatic Identification of Reactive Lymphocytes in Peripheral Blood

⭐ This code has been completely released ⭐

⭐ our article ⭐

If our code is helpful to you, please cite:

@article{mei2024adversarial,
  title={Adversarial training collaborating hybrid convolution-transformer network for automatic identification of reactive lymphocytes in peripheral blood},
  author={Mei, Liye and Peng, Haoran and Luo, Ping and Jin, Shuangtong and Shen, Hui and He, Jing and Yang, Wei and Ye, Zhiwei and Sui, Haigang and Mei, Mengqing and others},
  journal={Biomedical Optics Express},
  volume={15},
  number={9},
  pages={5143--5161},
  year={2024},
  publisher={Optica Publishing Group}
}

Requirements

Install PyTorch

# CUDA 10.2
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=10.2 -c pytorch

# CUDA 11.3
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge

Install remaining packages

pip install -r requirements.txt

Train

1. Prepare training data

The download link for our dataset is here.

PBC
├── train
│   ├── e
│   │   ├── 0001.jpg
│   │   ├── 0002.jpg
│   │   ├── .....
│   ├── l
│   ├── m
│   ├── n
│   ├── nbl
│   ├── vl
├── val
├── test

2. Weights (Optional)

The download link for our weights is here.

3. Begin to train

bash train.sh --model nextvit_small --batch-size 32 --lr 3e-4 --warmup-epochs 0 -weight-decay 1e-8 --epochs 100 --sched step --decay-epochs 80 --input-size 224 -resume ../checkpoints/nextvit_small_in1k_224.pth --finetune --data-path
your_imagenet_path

Test

1. Begin to test

bash train.sh --model nextvit_small --batch-size 32 --lr 3e-4 --warmup-epochs 0 --weight-decay 1e-8 --data-path your_dataset_path --resume ../checkpoints/your_checkpoints_path --eval

Results

1. Validation set

Methods	P(%)	R(%)	F1(%)	Images
Eosinophil	98.04	83.33	90.09	60
Lymphocyte	88.71	93.83	91.20	519
Monocyte	92.08	90.57	91.32	244
Neutrophil	98.28	99.59	98.93	975
Blast	92.59	87.55	90.00	257
Reactive Lymphocyte	93.17	90.17	91.54	447

2. Test set

Methods	P(%)	R(%)	F1(%)	Images
Eosinophil	97.67	91.30	94.38	46
Lymphocyte	90.22	92.06	91.13	491
Monocyte	94.00	89.10	91.48	211
Neutrophil	98.95	99.16	99.05	947
Blast	93.63	89.02	91.26	264
Reactive Lymphocyte	84.65	88.03	86.31	376

Acknowledgements

This code is built on Next-ViT. We thank the authors for sharing the codes.