Helicobacter Pylori Infection Classification Based on Convolutional Neural Network and Self-Supervised Learning

In this paper, a computer-aided diagnosis (CAD) method based on self-supervised learning was proposed for helicobacter pylori (HP) infection classification. The proposed method is composed of an encoder and a prediction head. The encoder can be trained by using self-supervised learning and contrastive loss. After obtaining the trained encoder, the prediction head can be trained by using the small medical image dataset. To evaluate the performance of the proposed method, some medical images are collected for testing. According to experimental results, the F1-score rates of the CAD system based on VGGNet-16 are 0.89 and 0.9 for HP+ and HP- images, respectively. The results show that the proposed method composed of VGGNet-16 and a multi-layer neural network can distinguish HP+ images from HP- images well. Compared with ResNet-50 and InceptionV3, VGGNet-16 can achieve a better classification performance. The experimental results show that VGG-16 can extract useful features from endoscopic images for HP infection classification via self-supervised contrastive learning.