| Summary: | 碩士 === 東海大學 === 資訊工程與科學系 === 94 === This thesis combines six practical textural features in medical ultrasound (US) images, i.e. block difference of inverse probabilities (BDIP), block variation of local correlation coefficients (BVLC), auto-covariance matrix, spatial gray-level dependence matrices (SGLDM), gray-level difference matrix (GLDM) and neighborhood gray-tone difference matrix (NGTDM) to classify breast tumors as benign or malignant. Firstly, the six textural features from an US image are performed as a textural feature vector. In general, the textural features always perform as a high dimensional vector; high dimensional vector is unfavorable to differentiate breast tumors in practice. Thus, the principal component analysis (PCA) is used to reduce the dimension of textual feature vector and then the image retrieval technique was utilized to differentiate between benign and malignant tumors. The image retrieval technique can reduce the training process when new cases were added. Moreover, the influence of variation caused by different US machines can be minimized by using this image retrieval method. The US dataset used in this study composed of 1020 sonograms of region of interest (ROI) subimages from 255 patients. Two-view sonogram (longitudinal and transverse views) and four different rectangular regions are utilized to analysis each tumor. The simulations demonstrate that the proposed computer-aided diagnosis (CAD) systems differentiate solid breast nodules with a relatively high accuracy and helps inexperienced operators avoid misdiagnosis.
|
|---|
