fyb99/DeepSNN

A large-scale dataset and a deep sequential neural network for IgA nephropathy histopathology identification and quantification

• This is the official repository of our paper entitled "A large-scale dataset and a deep sequential neural network for IgA nephropathy histopathology identification and quantification".

1. Environment

Python 3.7
PyTorch = 1.5.0
opencv-python

You can install them by the requirements.txt

pip install -r requirements.txt

2. Dataset

The dataset will be uploaded to Baidu Disk (LINK:https://pan.baidu.com/s/1uH9qmjYKhGvIOPjiXv2v_A). You should ask for the PASSWORD to download it by sending emails to pansai301@126.com. The email should contain your institution, your grade/occupation and the purpose of pursuing this dataset. The act of spreading our dataset freely on the Internet is forbidden.

3. Model

You can Download the pytorch model from BaiduPan(key: BuAa) and put the .pth files in the repository folder.

4. Test

1. Clone this repository:

   https://github.com/fyb99/DeepSNN.git
   

2. Place the test images in ./data/. For example, ./data/test_sample .

   data 
   └── test_sample
       ├── img.png
       └── mask
   

3. Run the following command for the IgA nephropathy histopathology identification task with the input whole slide image.

   python isolate_test.py
   

4. Finally, you can find the predicted results in ./data/test_sample/predicted.

5. Training

1. If you want to train the proposed method with our dataset or private dataset. Run the following command for training the lesioin segmentation subnet.

   python vanilla_train_seg.py
   

2. Run the following command for training the glomerulus classification subnet.

   python vanilla_cls.py
   

   Note how the training sets of two subnets are generated can be found in dataset.py and cls_dataset.py.