Medical CT Image Classification Based On Improved B-CNN Model

Medical imaging has brought rapid development to clinical diagnosis,and its imaging technology can provide intuitive diagnostic information for physicians,which is widely used in the diagnosis of various diseases.But the massive growth of medical image data has brought great pressure to related medical personnel.Computer aided diagnosis technology based on medical image can make up for the shortage of doctors,lack of experience,excessive fatigue and other problems,and has been rapidly developed in recent years.Computer aided diagnosis technology aims to distinguish patients’ symptoms and provide diagnostic information by subtle focal features in medical images,and its classification task objectives are consistent with fine-grained image classification algorithms.Therefore,this paper constructs a computer aided diagnosis system for medical CT images by using the weak supervised fine-grained image classification algorithm.By changing the original network classifier to improve the classification accuracy,adding dimension reduction algorithm to reduce the calculation and storage costs,and improving the model’s ability to detect the CT of COVID-19 patients through the classification weight assignment of the classifier,the main work is as follows:(1)The research background and significance of medical image classification technology were summarized.The fine-grained image classification task is the same as medical image classification,which is classified by the subtle features between different categories.This paper analyzes and summarizes the research status and development prospect of fine-grained image classification algorithm and the problems to be solved when applying fine-grained image algorithm to medical image classification.(2)A SVM-based B-CNN classification model was constructed and applied to medical CT image classification.Through lung CT image experiment,common convolutional neural networks such as Inception V3,Res Net50 and VGGNet were compared with B-CNN experiment.The feasibility of fine-grained image classification algorithm on medical images was verified by accuracy,recall rate and other indicators.The B-CNN network model selected with VGG16 and VGG19 as subnetworks has a good classification effect.The accuracy and recall rates reached 95.19% and 96.54% in the novel coronavirus CT image data set,and 98.38% and 97.86% in the pulmonary nodules CT image data set,respectively.The network classifier was improved to improve the classification accuracy of the algorithm.After softmax was replaced by SVM,the classification accuracy was increased by about 1.5percentage points and 2 percentage points respectively.Finally,the advantages of finegrained image classification algorithm were intuitively displayed through the visual algorithm.(3)The B-CNN novel coronavirus CT image classification algorithm based on GASVM was constructed.Aiming at the problem of too high dimension of double linear pooling feature graph,the dimensionality reduction algorithm of linear principal component analysis was selected to reduce the calculation and storage cost of model fitting through the experimental comparison of multiple dimensionality reduction algorithms.While ensuring accuracy,the classification time of test set was reduced from 35.80 s to 0.033 s,which effectively improved the classification efficiency of the model.At the same time,in view of the specificity of CT image diagnosis and the high cost of classification errors of samples infected with COVID-19,the sensitivity of the model to patients’ CT images was improved by category weight assignment.The recall rate of the model was increased from 95.44% to99.62%.