Detection and classification of gastric precancerous diseases using deep learning

nowadays, there are different types of endoscopic images due to the dramatic development of medical imaging technologies and according to the sensitivity of this issue, proper diagnosis and classification of different kinds of diseases is very important. Gastrointestinal tract of the patients can be observed by endoscopy and this technology can be used to diagnose gastric diseases. Gastric lesions are divided into three categories: Advanced Gastric Cancer (AGC), Early Gastric Cancer (EGC), and Gastric Precursor Disease (GPD). Early detection of these types of lesions can be helpful for cancer prevention and for this reason, detection accuracy is of great importance. Since endoscopic procedures are operator-dependent and human factors based processes, there is a high probability of error which by designing an intelligent system based on convolutional neural networks, different types of lesions and diseases can be classified more accurately. In the proposal system, the convolutional neural network model is trained by designing a convolutional neural network architecture and using data augmentation, dropout, noise and weight decay techniques. Using performance criteria such as accuracy, cost, and confusion matrix, the performance of this model is evaluated. Finally, the proposed model is tested on a dataset of 3673 images of gastric diseases such as Erosion, polyps and ulcers. The results show 89% classification accuracy of this model.