Computer-Aided Detection And Analysis Of Mammographic Mass

Early diagnosis and theropy of breast cancer is the only effective way to save lifes. Mammogram is the most preferred method for breast cancer census at present. However, the ambiguous characteristics of the early cancers and the subjective impact of the doctors will all probeblely induce the error and miss on diagnosing. With the rapid development of computers, computer-aidede detection of breast cancer has become one of the hot research points among medical image processing field. The effective computer-aidede detection method can help doctors realize and analyze the mammograms better, and further improve the accuracy of diagnosis. Mass is one of the major signs of breast cancer, which always takes on vague margin and very similar apperence with the normal tissues. Thus, it is a difficult and challenge problem of the computer-aidede detection system. The paper is established based on the previous research, and through incorporating the new ideas of computer vision and machine learning, to construct the computer-aidede detection and analysis system.(1) Morphological component analysis is introduced into the feature analysis of the mammograms for pre-processing. The representative sub-dictionaries should be incorporated into the morphological component analysis model, and then the mammograms can be decomposed as the piece-wise smooth component with major density distribution and the texture component with noises and vessels. It could provide better pre-processing result for the region of interest's (ROI') detection.(2) The coarse detection method based on growing styles of masses and adaptive threshold is proposed. The extension of multiple concentric layer method is conducted on the smooth component decomposed by morphological component analysis. Then following the characteristics of image histograms, the original constant threshold is replaced by adaptive one. It prevents the insufficient layer produced by the constant threshold and can effectively detect various mass regions.(3) A dual level set method and a vector-valued level set method based on relaxed shape constraints are proposed. Based on the ROIs detected by the coarse detection step, the level set method is introduced and extended to the dual level set one. With the shape constraint obtained by the first level set, the accurate segmentation results can be acquired by the second level set. For capturing radiated marginal characteristics of masses and further refine the marginal extraction result, the dual level set method is then extended to the relaxed shape constraint based vector-valued level set method. Through incoporating multiple image features, combining region-based and edge-based level set method, the better segmentation results can be obtained.(4) A pairwised constraint support vector machine based active learning scheme is proposed and applied to fine detection of mass regions. With the margin extraction results of mass regions, the characteristics within them can be analyzed. Then the restraint characteristics about extremely different masses of the same type and the similar masses from different types are considered. The pairwised constraint support vector machine with active learning mechanism is introduced for selecting samples with restrain characteristics from the detection results. Finally, the performance of the detection system is improved.(5) A mass retrieval scheme is proposed based on the Bag of Words model and multiscale partial pyramid matching scheme. Similar cases retrieval could provide good reference and help for the accurate diagnosis. Therefore, with the various types of masses detected by the fine detection procedure, the reginal characteristics are modeling by employing the Bag of Words model. The partial pyramid matching kernel function is then extended to the cross-matching one under multiresolution, and the matching ranking scheme can be established for searching the similar cases corresponding with the query mass regions.In this paper, we incorporate the computer vision and machine learning method into the medical image processing. The constructed computer-aided detection and analysis system for mammographic masses could effectively detect various types of masses and also find significant margins of different masses. The constructing of the system can effectively enrich the theory study of the computer-aided diagnose system.