Classification Of Liver Image Images Based On Saliency Algorithm

In recent years,medical image analysis has played an increasingly important role in the diagnosis of liver diseases.As an important branch of medical image analysis,medical image classification is used in image processing,pathological analysis,clinical diagnosis,surgical planning,and computer-aided diagnosis.This aspect has extensive research value.In order to improve the efficiency of liver image recognition,this paper combines the saliency algorithm with medical image recognition technology,and conducts research on different types of liver detection.The specific research content is as follows:Aiming at the problems of large individual differences in clinical liver tumors and low gray contrast with surrounding tissues.This paper proposes a dual-channel recalibration mechanism based on the attention module.First,the feature recalibration strategy is implemented for different channels,and new feature weights are given to them.Then it is embedded in the Inception ResNet V2 model to highlight the effective features,weaken the invalid features,and enrich the global features and detailed information of the image.In view of the current medical imaging equipment are all three-dimensional imaging,medical imaging data causes unnecessary waste and other problems.This paper proposes a strategy to combine the two-dimensional image domain and the three-dimensional space domain.First,rearrange the three-dimensional images,and then design a new TC3 D structure based on the good timing and robustness of the three-dimensional images.The dual convolutional pooling unit and nonlinear operation are used to extract the 3D features,then the same level of fusion is performed with the 2D network embedded with the dual channel recalibration mechanism,and finally the XGBoost classifier based on the integrated algorithm is constructed for the classification of liver images.Aiming at the problems of more redundant information in liver image slices and easy confusion of various organs and tissues.This paper proposes an algorithm for automatic extraction of target regions based on the human visual system.First,the saliency detection algorithm ResNet＿UNet is used to detect the target area of the image,and the ResNet structure is used as the coding structure of the saliency algorithm.The decoding part is composed of convolution blocks,batch operations and nonlinear functions.According to the generation of the binary saliency map,the Grabcut algorithm based on graph cut theory is used to iteratively converge and accurately locate the foreground area,and the foreground area is used as input to conduct experiments on the multidimensional network.This thesis uses a dual-channel recalibration mechanism to strengthen effective plane features,constructs a TC3 D network structure to extract temporal spatial semantic information,combines multi-dimensional features,combines saliency algorithms with deep learning,and achieves good classification results on different data sets.Prove the superiority of the algorithm in this paper.