Research And Implementation Of Liver Cancer Pathological Image Recognition Based On Multi-Space Feature Extraction

Liver cancer is a malignant tumor with high mortality,and it threatens human life and health directly.Early diagnosis can get more time for treatment and prolong the life of patients.Pathological images are the gold standard of liver cancer diagnosis.Combining Computer Aided Diagnosis(CAD)technology with liver cancer pathological diagnosis is of great significance to improve diagnosis rate of early liver cancer.We make the liver pathological image as the research object,and make the multi stages recognition of liver cancer as the research content.Because the single color space can not reflect the information contained in the image completely,we propose the method of mapping the original liver pathological image to different color spaces and texture spaces,and then extract features in these different spaces,which can obtain more comprehensive and useful classification information.In view of characteristics of experiment used liver pathological images,the mapping color spaces we selected contain R space,B space and Y space,and the mapping texture spaces we selected contain entropy space and LBP space.In feature extraction phase,because Local Binary Pattern(LBP)feature and Local Directional Pattern(LDP)feature sampling the neighbor pixels only by single scale,which can not obtain the information completely,and extend the sampling scale will increase the computation cost.We propose the Sparse Multi-scale Local Binary Pattern(SMLBP)feature and Sparse Multi-scale Local Directional Pattern(SMLDP)feature based on multi-scale sparse sampling method,which makes up the disadvantages of LBP feature and LDP feature.Because Higher Order Local Autocorrelation Coefficients(HLAC)feature only consider local information,and ignore the global information,which can not reflect the differences among pixels.We propose the Average Correction Higher Order Local Autocorrelation Coefficients(ACHLAC)feature,which makes up the above defects of HLAC.In establishment of classifier phrase,because each classifier has its advantages and its disadvantages,we combine several classifiers to establish classification model,and determine the classification result by voting,which can improve the classification performance.To verify the effectiveness of the algorithm,we make lots of contrast experiments,which verify the effectiveness of Grid Search Method to optimize SVM parameters,the effectiveness of multi-space mapping,the good classification performance of SMLBP feature,SMLDP feature and ACHLAC feature,and the availability of classification model by voting.The experiments indicate our method has 95.1%total classification accuracy for liver pathological images,which can classify the liver cancer pathological images primely and has a good application prospects.