Gastrointestinal Tract Disease Classification Using Residual-Inception Transformer With Wireless Capsule Endoscopy Images Segmentation

Abstract

One of the leading causes of cancer-related mortality in affluent nations is Colorectal Cancer (CRC). Gastrointestinal (GI) disorders pose a severe hazard to human health and are frequently managed through invasive operations. Accurately and early CRC diagnosis is crucial for the course of treatment. Experts can determine characteristics in the human GI tract using an endoscope to look for infections or even potential CRC indications. However, in the research done so far, the desired accuracy value could not be attained. In this study, the approach of segmenting GI tract regions in Wireless Capsule Endoscopy (WCE) scans was used together with a long-range transformer model. In order to produce local information with surrounding pixels and patches in the architecture, embedding a split token method is applied to the transformer block. To diversity model training effectiveness and capture intricate anatomical details of the GI tract, feature learning with a cross-channel technique was used along with the Residual block and Inception module. The proposed framework was tested with experiments on the WCE Curated Colon Disease (WCECCD) dataset, which contains a total of 4 classes of images, including 3 different GI tract diseases and normal. The proposed model reached 99.50% accuracy performance. The proposed approach outperformed current state-of-the-art techniques, according to experimental results. This transformer model, developed together with GI tract image segmentation, is making significant progress in the field and become a more precise and effective tool for experts.