Pancreatic cystic tumor CT image classification method based on multi-channel multiple classifiers

Abstract

A pancreatic cystic tumor image classification method based on multi-channel multiple classifiers comprises the following steps: 1) performing window width and window level adjustment on an original image, and performing Canny edge detection and gradient amplitude calculation to enhance edge features; 2) adopting a ResNet to perform end-to-end training on a multi-channel graph, using an output ofa pool5 layer as extracted features, using a Bayesian classifier and a KNN classifier to perform classification, and obtaining the classification probabilities; and 3) using a random forest classifierto classify the obtained 3 different probabilities to get a final result, wherein a random forest is composed of multiple decision trees. The pancreatic cystic tumor image classification method basedon multi-channel multiple classifiers can automatically perform edge enhancement, and can improve classification accuracy.

Description

technical field [0001] The invention belongs to the technical field of image classification, and in particular relates to an image classification method. Background technique [0002] Usually, patients need to undergo a series of examinations before undergoing surgery, among which the observation of organs and lesions through computed tomography (CT) imaging is an indispensable method. Currently, physicians use their experience to determine whether cystic tumors of the pancreas are serous or mucinous. Existing image classification methods can be divided into traditional methods and deep learning methods according to their characteristics. Traditional methods first manually segment the tumor area; then, use different feature extractors to extract features, such as: texture features, structural features, etc., and then classify them with a classifier. The method of deep learning also needs to segment the original image first, then use the convolutional neural network to extr...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the human error introduced by manual tumor segmentation and the insufficiency of a single classifier affecting the results

Images