Recognition of High-grade Squamous Intraepithelial Lesion

This repository includes the implementation of a classification system of HSIL (High-grade Squamous Intraepithelial Lesion) based on CNN architectures.

Introduction

High-grade squamous intraepithelial lesion (HSIL or HGSIL) indicates moderate or severe cervical intraepithelial neoplasia or carcinoma in situ. It is usually diagnosed following a Pap test. In some cases these lesions can lead to invasive cervical cancer, if not followed appropriately.1

Pap test will generate three types of images as follows:

We focus on classificaiton of the above three types of colposcope images from Pap test using the method of deep learning.

Requirements

• Python (2.7.14)
• Keras (2.0.5)
• tensorflow (1.4.0)

Manual

• Train

python train.py -net -save_dir

• Predict

python predict.py -weight_model -image_dir

References and papers

• Vgg19 Network    -  Very Deep Convolutional Networks for Large-Scale Image Recognition
• Residual Network
  • Deep Residual Learning for Image Recognition
  • Identity Mappings in Deep Residual Networks
• ResNeXt
     -  Aggregated Residual Transformations for Deep Neural Networks
• DenseNet    -  Densely Connected Convolutional Networks
• SENet    - Squeeze-and-Excitation Networks