A Study Of Computer Aided Diagnosis Of Breast Cancer

At present,mammography is the most effective method to diagnose mammarydiseases especially the breast cancer.The quality of mammogram is poor for the natureof breast tissue,so that the doctor missed the tiny lesion.A computer-aided imageanalysis technique can provide a consistent and reproducible 'second opinion'to aradiologist,which may reduce false-negative diagnosis.Clustered microcalcificationsare the most important indications of malignancy on mammogram. So the diagnosis ofbreast cancer can be done through the classifications of the clusteredmicrocalcifications.In this thesis, some key issues of the Computer Aided Diagnosis(CAD) techniqueof breast cancer microcalcifications are systematically investigated.Some newlyalgorithms have been adopted to detect the microcalcifications and classify theclustered microcalcifications.Firstly, a algorithm is presented based on the wavelet tosegment the exact microcalcifications regions;Secondly, we proposed Support VectorMachine(SVM) automatically to learn the relevant features as input to the SVMclassifier to reduce the false positive.Suspicious clustered microcalcifications arelabeled by the criterion whether three or more microcalcifications are detected withinan area of 1cm2 after grouping the microcalcifications. Finally,a pool of manyfeatures with the information about the texture,shape and so on of clusteredmicrocalcifications are extracted.The SVM classifier is used to label the clusteredmicrocalcifications as either malignant or benign.The experimental results with the real clinic image data demonstrate the effectivenessof the proposed algorithm.Because the detecting algorithms and training methods aregeneral,they can be applied to other small and dim targets detection problems.