WCE Images Polyp Segmentation System Using Convolutional Neural Network (CNN) With Stochastic Gradient Descent Optimizer

Abstract

Polyps in the small bowel have a chance of developing into cancerous tumors. As a result, it is important to recognize and treat such polyps at an initial stages. This would significantly boost the patient's chance of survival. Due to the rapid advancement of technology, wireless capsule endoscopy is regarded as a medical breakthrough. This allows for easy, painless, and inexpensive observation of the interior body, which is not visible to the naked eye. Simultaneously, the wireless capsule endoscopy's low-quality images are considered as its primary weakness. As a result, certain forms of polyps cannot be diagnosed from this wireless endoscopic imaging, even by a highly qualified physician. As a result, computer-aided polyp identification remains an ongoing challenge. This research introduces a novel segmentation algorithm for this purpose. The purpose of this research is to present a modified convolutional neural network (CNN) algorithm for wireless capsule endoscopy image segmentation that is based on dropout and the stochastic gradient descent optimizer. To increase feature extraction accuracy while decreasing time costs, this work analyses the CNN structure, the over fitting problem, and the combination of dropout and the SGD optimizer with the CNN. Additionally, this novel innovation was assessed using many polyp databases and its experimental results were compared to those of previously developed polyp segmentation techniques. The results demonstrate that our enhanced CNN outperformed state-of-the-art techniques.