Jiaxin-Ye / Aided-Diagnosis-System-for-Cervical-Cancer-Screening

Aided-Diagnosis-System-for-Cervical-Cancer-Screening

Journal Link | Quick Start | Cite

Recommend the top k lesion cells and predict the positive probability of WSIs.

---

Requirements

Hardware

GPU: Nvidia 1080Ti or better (at least 10G memory)
CPU: Intel i7 or better
System Memory: 16G or better

Software

System: Win10
Nvidia GPU corresponding driver
CUDA: cuda 9.0
cudnn: cudnn 7.0

Python

Python: 3.6
Tensorflow-gpu: 1.7.0
Tensorboard: 1.7.0
Keras: 2.1.2
Keras-Applications: 1.0.6
Numpy: 1.19.5
Openslide-python: 1.1.1
Opencv-python: 3.4.1.15
Pandas: 0.20.3
Scikit-image: 0.17.2
Scikit-learn: 0.23.2

---

Supported WSI formats

WSI formats supported by the opensource OpenSlide library, including x.svs, x.mrxs, x.tif, etc; WSI resolution: 20× or 40× （0.1 – 0.6 um/pixel, 0.1 – 0.4 um/pixel is better)

---

Quick Start

Installation

Step1. Install CUDA v9.0 and cuDNN v7.0.5

Step2. Download Aided-Diagnosis-System-for-Cervical-Cancer-Screening

git clone [e-mail]:ShenghuaCheng/Aided-Diagnosis-System-for-Cervical-Cancer-Screening.git
cd Aided-Diagnosis-System-for-Cervical-Cancer-Screening

Step3. Install requirements

pip3 install -U pip && pip3 install -r requirements.txt

Train

Train Model1

# train model1 classifier
python tools/train.py -n model1-cls -b 16
# train model1 locator based on model1 classifier's backbone
python tools/train.py -n model1-loc -b 16 -w [path to model1-cls weight]

Train Model2

# train model2 classifier
python tools/train.py -n model2-cls -b 32

Train WSI Classifier

# train WSI classifier
python tools/train.py -n wsi-cls -b 64

Evaluation

Evaluate classifiers.

python tools/eval.py -n model1-cls -b 16 -w [path to evaluated weight]
                        model2-cls -b 32
                        wsi-cls    -b 64

Inference

Python

Do inference to WSIs according to config file.

python tools/inference.py -c configs/wsi_inference.py -f [path to WSI or path to WSI list files] [--intermediate]

C++ Software

Prepare: convert h5 weights to pb files.

python tools/convert_to_pb.py -m model1 -w [path to weights] -o [path to save]
                                 model2
                                 wsi_clf_top10
                                 wsi_cls_top20
                                 wsi_clf_top30

Do inference: see C++ software

Tutorials

• For dataset config file, see config for dataset
• For train and eval config file, see configs
• For inference config file, see configs

---

C++ software

The C++ software with test WSIs is available at Baidu Cloud. Correspongding user manual pdf is uploaded.
Extracting code can be provided by email [e-mail] or [e-mail].

---

Reference

If our work is useful for your research, please consider citing our paper:

Cheng, S., Liu, S., Yu, J. et al. Robust whole slide image analysis for cervical cancer screening using deep learning. Nat Commun 12, 5639 (2021). https://doi.org/10.1038/s41467-021-25296-x

@article{cheng2021robust,
  title={Robust whole slide image analysis for cervical cancer screening using deep learning},
  author={Cheng, Shenghua and Liu, Sibo and Yu, Jingya and Rao, Gong and Xiao, Yuwei and Han, Wei and Zhu, Wenjie and Lv, Xiaohua and Li, Ning and Cai, Jing and others},
  journal={Nature communications},
  volume={12},
  number={5639},
  year={2021},
  publisher={Nature Publishing Group}
}