Study Of Wavelet-based Methods For Mass Segmentation In Mammograms

Early diagnosis and treatment of breast cancer is an efficient way of reducing themorality of breast cancer. Observation of abnormalities through mammograms is thepreferred method for breast cancer census. With the rapid development of imageprocessing technology, Computer Aided Detection (CAD) becomes more and moreimportant in breast cancer detection. The thesis introduce the author’s work ondeveloping a CAD system for mammograms, mainly including image preprocessing,extraction of region of interest and edge detection of breast mass.The proposed method firstly preprocesses the X-ray image using morphologicalprocessing, region growing and other related image processing technology. Then weput forward a new method to segment the mass in a mammogram. The proposedmethod gets the gray-level histogram of the X-ray image at first, and then usingwavelet transformation to find the Wavelet Transform Modulus Maxima. Accordingto this modulus maxima, the coarse segmentation threshold can be found and thesuspected regions in the X-ray image can be detected. After this rough segmentation,region growing method and local-based active contour model are used to get the edgeof breast mass.There are65mammography images from MIAS database chosen as test imagesand there is one mass in each image at least. Use the proposed method to test the65images and compare the results with the experts’ marking information contained inthe database. The detection rate is95.5%and the average false positive is0.84perimage when we use Daubechies wavelet. The proposed method can effectivelysegment the masses in mammograms with a high detection rate. It’s proved to have ahigh detection rate and further research and application value.