Medical Image Classification Based On Autoencoder Dimension Reduction And Similarity Measurement

The COVID-19 has had a huge impact on the world economy and society.The imbalance of distribution in medical resources and the shortage of medical staff have put tremendous pressure on the medical system.With the development of computer technology,the application of artificial intelligence technology for computer-aided diagnosis is an important development trend.It can not only reduce the pressure of medical staff,but also improve the efficiency of disease diagnosis and treatment.In addition,the development of computer hardware and the substantial increase in computing power have made it possible to apply deep learning technology for computer-aided diagnosis.Meanwhile,the crossing and blending of computer technology and medicine provide new opportunities for research and clinical practice in the field of computer-aided diagnosis.Therefore,significant progress was made on deep learning technology in recent years.Some scholars have made further research and applied it to the field of medical image analysis.Compared with traditional manual feature extraction methods,deep learning can conduct autonomous learning through a large amount of data without human intervention.In the medical image classification system,the fine structural features of the medical image extracted by the deep learning network will directly affect the performance of the model.Although deep learning has made good achievements in medical image classification,there are still some problems that need to be solved urgently in the field of medical image processing.This article focuses on solving the following problems: First,the deep learning network processing high-dimensional medical image data needs to calculate a large number of parameters,and the long network training time may lead to the occurrence of dimension disaster.Second,the original medical images usually lack precise labels.Therefore,the professionals have to spend a lot of time and effort in labeling medical images.In the case of a small number of labeled medical image samples,how to use limited image data to efficiently train and optimize the deep learning model to obtain satisfactory results is also one of the current challenges.Based on the full absorption and reference of previous research results,this paper applies a mixed order stacked autoencoder to unsupervised dimensionality reduction of medical images for the first problem,and solves the problem of dimensionality reduction of unlabeled high-dimensional medical images.By studying the deep learning autoencoder network,the autoencoder network is used to reduce the dimensionality of medical images in an unsupervised manner,and the features of the reduced-dimensionality image are combined to construct a mixed order feature matrix,and the mixed order feature matrix is used for medical image classification.Medical image classification algorithm based on stacked autoencoder.The experimental results show that the proposed algorithm can better reduce the dimensionality of medical images compared with the six comparison algorithms,and obtain better classification accuracy and shorter classification time as well.When it comes to the second problem,feature activation visualization and image similarity measurement methods are used to efficiently migrate,train and optimize the depth model,which effectively reduces network hyperparameter adjustments and reduces the dependence of network training on label images.Finally,the optimized network is used to extract features from the CT image,and the K-nearest neighbor algorithm,support vector machine and random forest are used for classification and evaluation.Experiments are performed on the public CT image data set with six comparison algorithms.Experimental results show that the proposed algorithm has the smallest model,the least classification time,and a better robustness.