Classification of Gastric Precancerous Diseases using Hybrid CNN-SVM

Abstract

Gastric cancer (stomach cancer) is now the sixth most common diagnosed cancer and the third leading cause of cancer mortality in the world. Gastric Erosion, Gastric Ulcer, and Stomach Polyp are examples of Gastric Precancerous Diseases (GPDs) that can lead to gastric cancer if not recognized early or misdiagnosed. Classifying these GPDs is a difficult task. Undoubtedly, Deep learning networks (DNNs) have shown to be effective in solving the challenge of image categorization. Next to practical difficulty is the limitation of the availability of medical images for DNN training. In this paper, a hybrid model is proposed to classify GPDs. The model is a combination of Convolution Neural Network (CNN) Gastric Precancerous Diseases Feature Extractor Network (GPDFENet) for feature extraction and Support Vector Machine (SVM) for classification. An open dataset “Data-Open-Access4PLoS-One” including erosion, ulcer, and polyp endoscopic images were utilized to train the network. After evaluation, the network is then compared to various pre-trained networks such as AlexNet, ResNet-50, ResNet-101, and Inception V3. The proposed model (GPDFENet+SVM) has achieved an accuracy of 93.22%.