| Summary: | 碩士 === 義守大學 === 資訊工程學系 === 106 === Mammography is widely clinical tool in screening breast cancer and the density of breast gland is highly related with breast cancer. However, the determination of densification of breast image was by visualization and experience. Hence, the classification between normal and abnormal cases was using features of image in this study.
The two-groups retrospective study was involved in this study by collected mammography with reference of BI-RADS. The experimental and control groups were collected numbers of sample 640 and 324 with respectively. The region of interest (ROI), thresholding and edge detection were utilized to generate a binary image which was transformed to gray level co-occurrence matrix (GLCM). The features were extracted from both original and GLCM images in order to apply classified study. The Mann-Whitney test was applied to exam the significant differences between physiological measurements and textural features. The logistic regress (LR) and support vector machine (SVM) were adopted to perform classification.
The features were extracted from both original and GLCM images were shown significant difference between groups by Mann-Whitney testing (P<0.05). The textural features extracted from GLCM with angle 135 degree were classified by SVM. The sensitivity, specificity, PPV,NPV, AUC, accuracy and Kappa were shown 67.3%, 97.2%, 92.4%, 85.4%, 82.2%, 87.2% and 0.691.
The performance of classification by SVM was superior than those by LR. The features of mean, SD, entropy and skewness extracted from original images and contrast, correlation, energy and homogeneity extracted from GLCM could obtain feasible classification by SVM.
|
|---|
