The human gastrointestinal (GI) system is vulnerable to various diseases, such as inflammatory bowel disease, peptic ulcers, and gastroesophageal reflux disease. Endoscopy is the conventional method used to assess the gastrointestinal tract. An ordinary endoscopy session necessitates the identification of several diseases, which is a monotonous and lengthy process to complete manually. Computer-aided diagnostic systems have been employed to help doctors analyze data. However, they may need a more refined technique due to the limitations of feature extraction procedures. Convolutional neural networks (CNNs) reveal promise but have necessitated extensive resources, making them inappropriate for real-time diagnosis in resource-constrained conditions. This article has presented GastroNet, an efficient approach to classifying stomach cancer using endoscopic images. GastroNet utilizes ShuffleNet to extract features and a Feature Pyramid Network to handle images of varying sizes. GastroNet is then combined with NanoDet for detection and classification. The outcomes on the Hyper Kvasir dataset demonstrate that GastroNet achieves a mean accuracy of 97.3%. This high accuracy rate signifies excellent functionality in comparison to other deep-learning-based models. Additionally, GastroNet operates rapidly by processing videos at 97 frames per second, enabling real-time clinical diagnosis.
Raza et al. (Mon,) studied this question.