Research On Image Recognition Algorithm Of Colon Cancer Based On Texture Features

Colon cancer is one of the most common malignancies in the world,leading to about 50 million deaths each year,and the incidence increasing year by year.However,compared with other cancer,the cure rate and 5 years survival rate is relatively high.The survival rate was directly related to the diagnosis period,and histopathological examination was the most common diagnostic method.A large number of studies have shown that computer-aided diagnosis techniques can improve the accuracy of diagnosis of colon cancer.In this paper,we study the pathological image recognition algorithm through two aspects: segmentation and classification.This paper is organised as follows:1.An algorithm for gland segmentation based on object structure has been proposed.With the help of object-oriented technique,the region of the lumen is used to initialize the seed points by the circle fitting method.The region of the nucleus is used to find the boundary of the region growing.Because the color of the nucleus is different from other tissue in image,the speckle detection algorithm based on Lo G is used to identify the location of the nucleus.According to the distribution of the nucleus,the histological features of the pathological images of colon cancer were obtained.2.A method for identifying colon cancer images based on multiple features has been proposed.This paper combines the characteristics of HOG,GLRLM and histological characteristics.At the same time,we uses feature selection algorithm to reduce the impact of irrelevant features considering the redundancy of the information contained in different feature sets.Finally,the feature set of the feature selection is normalized.The parameters of the classifier are set by the method of grid search.The experimental results show that the recognition method based on multiple features can improve the recognition accuracy.