Research On Aided Diagnosis Algorithm Of Microcalcification Based On Wavelet

Breast cancer is the most frequently diagnosed cancer in women. Early detection and diagnosis represent a very important factor in breast cancer treatment and consequently the survival rate. Digital mammogram is considered to be the most reliable method of early detection of breast cancer. As its visual clues are subtle and varied in appearance, microcalcification detection and diagnosis is a challenging work for specialists. The computer aided diagnosis systems have been developed to aid radiologists in microcalcification detection and diagnosis. Currently, the performances reported in the literature are better for microcalcification detection than diagnosis. And the diagnosis results can't meet the clinical needs.Wavelet transform have been proved to be effective in classification of benign and malignant microcalcification. Little attention is paid to the selection of wavelet basis and its effect on feature extraction in current applications based on wavelet in microcalcification diagnosis. Moreover, the common features based on wavelet are too simple to get a satisfied classification results.In this paper, we make a research on characteristics of wavelet basis and its effect on feature extraction. And we adopt the scalar wavelets, multi-wavelets, directional-wavelets and dual-tree complex wavelets to extract the muti-level information in mammogram. Two effective feature sets are proposed for feature extraction. An aided-diagnosis algorithm based on wavelet, combining with Genetic Algorithm(GA) and k-nearest-neighbor(KNN) classifier is proposed.Receive Operating Characteristic (ROC) curve and Leave-one-out method are used to evaluate the performance of our proposed algorithm. The experimental results shown that the proposed algorithm can produce a high classification rate. Validated by the same mammographic database-Nijmegen, our algorithm is superior to previous methods. Some reasonable suggestions are also presented through analysis on experimental results.