Research On Feature Classification Recognition Of Medical CT Images Based On Convolutional Neural Networks

In modern medicine,medical imaging is a technology that generates digital images through medical imaging equipment and has a wide range of applications in medical diagnosis,treatment planning,disease research,and other areas.In recent years,with the rise of machine learning image processing technology,new approaches have been provided for the processing of medical images.However,medical images present challenges due to their large information content,high complexity,small difference in image appearance,location of lesion area and data sample balance.When traditional algorithms are used to solve these problems,there are issues such as weak feature extraction ability,inaccurate localization of lesion areas,and poor generalization ability.Therefore,this thesis proposes a medical image classification and recognition processing method based on the Convolutional Neural Network(CNN),which is efficient,high accuracy and has strong generalization capability.The research contents of this thesis include:(1)A medical CT image classification and recognition algorithm based on synchronous convolution and cross-fusion is proposed,which can effectively address the challenges of difficult feature extraction and inaccurate positioning of lesion areas in medical images.The algorithm takes into full consideration the information of lesion areas and conducts differential learning in a targeted manner.For the experimental results,this thesis adopts multiple validation methods and conducts experiments on a publicly available COVID-19 CT dataset.The results indicate that the algorithm has better classification sensitivity and specificity performance while ensuring model complexity.(2)A multi-level feature aggregation classification algorithm based on CNN is proposed.Building upon the algorithm presented in Chapter 3,this optimization addresses the issue of minimal variations in target appearance by performing multi-level feature fusion on different tiers of feature maps,while maintaining a balance between the complexity of the network model and the size of the dataset.Through comparative experiments,compared the performance of different optimizers and selected the optimal optimizer to apply to the algorithm model,verifying the effectiveness and overall superiority of the algorithm.The final experimental results demonstrate that the multi-level feature fusion classification algorithm effectively improves both the training performance and generalization ability of the model.This research is based on the CNN algorithm to address the problems of high complexity and technical complexity in medical image processing models.In the feature extraction stage,the proposed model takes full consideration of the lesion area feature and conducts targeted differentiated Learning.During feature fusion phase,the algorithm model proposed in this thesis effectively performs feature map fusion at different levels.The ultimate experimental results demonstrate that the CNN based algorithm exhibits a good efficacy in solving the issue of medical image feature classification.These technologies have enhanced the level and efficiency of healthcare,allowing physicians to diagnose and treat diseases more accurately.