Research On Gleason Intelligent Grading Of Pathological Prostate Images Based On Image Enhancement

The correct diagnosis of prostate cancer is the primary prerequisite for effective diagnosis and treatment.The traditional diagnosis and treatment methods have high professional requirements for pathologists,and manual diagnosis and treatment is timeconsuming and low repetitive.In order to reduce the probability of misdiagnosis and missed diagnosis in the process of diagnosis and treatment,and improve the treatment efficiency of cancer.In this thesis,the deep learning algorithm is applied to the analysis of prostate pathological images.Computer-aided system can provide objective diagnosis basis and help doctors to make a more accurate diagnosis.The glandular structure of prostate pathological images is very complex,especially there is little difference in the degree of gland differentiation between Gleason grade 3 and Gleason grade 4.Therefore,misdiagnosis often occurs in the identification of Gleason3 grade and Gleason4 grade pathological tissue.In addition,there are few prostate pathological images marked by professional pathologists,which can not meet the large-scale data requirements of network training.Therefore,this thesis proposes a study on Gleason intelligent grading of prostate pathological images based on image enhancement.The main contents are as follows.(1)In the process of data preprocessing,the traditional geometric transformation is used to expand the data.Combined with the gray histogram of the image,the image data that meet the standard is screened out.Then the expanded data is enhanced by homomorphic filtering.It makes the morphological features such as texture and size of cells and tissues in pathological images become clearer,and provides more abundant image information for subsequent feature recognition.By adopting the method of image enhancement,the over-fitting phenomenon of the model due to the lack of data can be avoided and the classification performance of the model can be improved.(2)A Gleason grading model for four classification of prostate pathological images based on Alex Net network is proposed.Three feature extraction modules are introduced into the network architecture of Alex Net,which are dual-channel input module,dense residual connection module and RCAB module.This makes the context information of each feature extraction layer related to each other,and improves the feature extraction ability of the model,which is beneficial to the prediction of cancer grade.The experimental results show that the consistency between the prediction results of the model and the score marked by pathologists is 0.801,which proves that the model can achieve better Gleason classification of prostate pathological images.(3)A Gleason classification model of prostate pathological images based on VGG16 network is proposed.The dual-channel input method is adopted in the model,and three different Inception structures are introduced on the basis of VGG16 network for multiscale feature fusion to achieve different levels of feature extraction.The experimental results show that the recall rate of the model is 92.9% for the pathological images with Gleason score of "3 to 4",and the pathological images with Gleason score of 7 are classified with high accuracy.