Research On Medical Image Classification Based On Feature Selection And Its Application

Along with the continuous development of the computer vision technology and the medical imaging equipment,the information that medical image contains is extremely rich,so the classification technology of medical image is more and more important,and the extraction and selection of image characteristics are the key points of image classification,both domestic and foreign scholars are more concerned.With the improvement of the image processing technology,the dimension of the image information is more and more high,so it is very important to select the more effective features to classify the image.This paper mainly studies the medical image of rheumatoid arthritis(RA),taking the optical coefficients as the feature are extracted from Diffuse Optical Tomography of finger in patients with RA,other than the previous medical image feature extraction mostly based on texture,color and shape of image.Through the experiment of RA image classification,the feasibility of the optical coefficient as the classification feature is verified?Feature selection is a key method for medical image diagnosis,comparing many feature selection methods for optimization,the feature selection algorithm of Maximal Relevance and Minimal Redundancy(MRMR)is studied in detail.This algorithm combines the correlation and redundancy of the features,and uses the entropy operation to sort the features to select the best feature combination.Further research found that the MRMR algorithm has different weights in the aspect of correlation and redundancy,and the ranking of features will be different,so we investigate the adding of the weights of feature,using the feature sequences in descending order,then the sequence of features under different weights is sorted again according to the weight coefficient of the location so as to get the final feature sequence.By the modified feature sequence,heart disease and RA diagnostic classification experiments have been done and the results prove that the modified MRMR algorithm is effective in classification experiments.Before the medical image classification,in order to improve the accuracy rate of the classification,it is necessary to deal with sample selection,and the speed of disease diagnosis is very important to the patient,so it is important to seek a small sample learning method.Support vector machine is the best suitable learning method based on Structural Risk Minimization Theory for small sample learning,having good generalization ability,and can decrease the error of the training.However,in the process of establishing the classification model,there are some problems in parameter and kernel function selection.In this paper,the clustering algorithm based on density is used to combine particle swarm optimization to optimize the classifier,streamline the various sets of data and improve the accuracy of classifier,the simulation results show that the method is effective,and has the reference value for the medical image processing that needs reasonable selection.