Analysis of the shape and texture of breast masses in mammograms
MetadataShow full item record
AbstractThis thesis demonstrates the use of fractal analysis to characterize the shape and gray-level complexity of breast masses for their classification as benign or malignant. The fractal dimension (F D) computed using the ruler method applied to signatures of contours of masses resulted in the area under the receiver operating characteristic curve (AUC) of 0.89. F D with the shape feature of fractional concavity resulted in AUC = 0.93. The blanket method to compute F D related to gray-level variations provided poorer results. The results indicate that fractal analysis is more suitable to characterize the shape than the gray-level variations of breast masses. The effect of pixel size on texture features based on gray-level co-occurrences was also investigated. The texture features computed with the pixel resolution of 200 micrometers provided the highest AUC. The methods presented in this thesis are useful for computer-aided diagnosis of breast cancer.
Bibliography: p. 112-124