A two-stage method for polyp detection in colonoscopy images based on saliency object extraction and transformers

The gastrointestinal tract is responsible for the entire digestive process. Several diseases, including colorectal cancer, can affect this pathway. Among deadliest cancers, cancer second most common. It arises from benign tumors in colon, rectum, and anus. These tumors, known as polyps, be diagnosed removed during colonoscopy. Early detection essential to reduce risk of cancer. However, approximately 28% polyps are lost examination, mainly because limitations diagnostic techniques image analysis methods. In recent years, computer-aided these lesions have been developed improve quality periodic examinations. We proposed an automatic method polyp using colonoscopy images. This study presents a two-stage images transformers. first stage, saliency map extraction model supported by extracted depth maps identify possible areas. stage consists detecting resulting combined with green blue channels. experiments were performed four public datasets. best results obtained task satisfactory, reaching 91% Average Precision CVC-ClinicDB dataset, 92% Kvasir-SEG 84% CVC-ColonDB dataset. demonstrates that efficiently combination maps, salient object-extracted