Identification Of Colon Pathology Images Based On Deep Learning

Recently,the research of "smart medicine" based on deep learning has been widely concerned by researchers.Due to the continuous improvement of basic medical facilities,the explosive growth of medical data and the increasing number of patients,the methods of manual diagnosis relying on doctors and computer-aided diagnosis no longer meet the needs of development.Therefore,full automatic intelligent diagnosis has become a new research hotspot.Colon cancer identification is of great significance in medical diagnosis.Real-time,objective and accurate inspection results will facilitate medical professionals to explore symptomatic treatment promptly.However,it remains a great challenge for the fully automatic cancer diagnosis which is completely independent of doctors,especially the intelligent cancer diagnosis which relies on medical image analysis.Regarding the issue above,we propose a multi-scale feature fusion convolutional Neural network(MFF-CNN)based on shearlet transform to identify histopathological image of colon cancer.The network first uses shearlet transform to perform multi-scale and multi-directional transformation on colon cancer histopathological images and obtain shearlet coefficients.Then the shearlet coefficients of histopathological image in multiple decomposition scales were extracted as supplementary feature which were also fed to the network with the original pathological image.After feature learning and feature fusion,the MFF-CNN based on shearlet transform can achieve the identification accuracy of 96% and average F-1 Score of 0.9594 for colorectal adenocarcinoma epithelium(TUM)and normal colon mucosa(NORM).The superior performance of the network provides effective means for real-time,objective and accurate diagnosis of cancer.In practical applications,not all information is valuable for image recognition.We are more inclined to obtain key and effective information to speed up network recognition.At present,based on the characteristics that the visual system tends to focus on the critical information in the image and ignore the irrelevant information,researchers propose a series of attention related methods,hoping to dynamically focus on some parts related to the target task and suppress useless information,so that more computing resources can be applied to critical information,and more accurate image recognition can be achieved.Therefore,this paper combines the attention mechanism and convolutional neural network to realize the recognition of colonic histopathological images.Firstly,convolution neural network is used to extract the preliminary information from colon tissue image to generate feature map.Secondly,the attention mechanism module is embedded convolution neural network to make the model pay attention to the intermediate feature map along the channel and spatial dimensions,and then extract effective information.Finally,realize the identification of colon tissue through feature fusion.The model can subdivide the features of the intermediate feature map,and then obtain more discerning features.Experimental results show that this method can effectively improve the recognition accuracy of colon histopathology images.