CNN and Data Mining Based Endoscopic Ultrasound Image Recognition

Abstract

Therapeutic staff must repeatedly observe and compare endoscopic ultrasound images, which have the individuality of changing images and irrelevant gray-scale changes. A system proposal that is fit for image processing is proposed in light of the aforementioned uniqueness of ultrasound imaging. It can analyze the biliary region, gallbladder, abdominal lymph nodes, liver, descending duodenum, duodenal bulb, stomach, pancreas, and pancreatic lymph nodes, with a total of 10 ultrasonic organs, 21 types of sub-categories, and 3498 images. Binarization, histogram (HS) equalization, median filtering, and edge enhancement algorithms are used to preprocess the images. The data set is trained with the enhanced YoloV4 convolutional neural network (CNN) algorithm, and high precision is detected in real time. At long last, the normal precision of this calculation has reached 91.59%. This paper's algorithm has the potential to make up for the shortcomings of the original image detection system's manual detection, increase detection efficiency, reduce detection errors, and encourage the development of automated and intellectual detection in the health field.