Research On Mass Feature Extraction And Feature Transformation In Mammography

Breast cancer has become one of malignant disease which threatens the health of global women. The incidence and mortality of breast cancer are rising worldwide, and present a trend of younger age due to the living habits, work stress, social environment and other issues. Early discovery and early treatment can reduce the cost of treatment and improve the chance of survival effectively. In order to reduce the workload of doctors and improve clinical detection accuracy, computer aided diagnosis system has been applied to clinical diagnosis. Mass is a common symptom of breast cancer and one of the main functions of the computer aided diagnosis system is to distinguish benign lesions and malignant ones. However, because of the complexity of the shape and edge of the mass which is often associated with the surrounding tissues, the image is often accompanied by noise and low contrast which bring a great challenge to the mass classification.Effective feature extraction method is the key to obtain accurate classification results. This paper proposes new feature extraction methods in the aspect of spatial information, semantic information and lexical weight based on bag of words model. The main work of our research is summarized as follows: a feature extraction method based on anisotropy edge annular region feature extraction method is proposed, which highlights the importance of edge information in the mass classification. The global information and local information of the image are obtained at the same time. A feature fusion method based on semantic similarity is proposed through which the extracted feature contains both the semantic information and spatial information of the image. Moreover, the algorithm complexity is reduced and operation efficiency is improved. A feature transformation method based on Dirichlet Fisher kernel is proposed since the visual words in bag-of-words feature is low discrimination ability. The transformed feature is some sort of the TF-IDF weighting method which improves the weight of important visual words and reduces the weight of unimportant words. As a result, the feature discrimination ability is improved effectively.The experiment results show that the proposed feature extraction and transformation methods can express spatial structure and semantic information of the mass effectively. The methods we proposed improve the classification accuracy and provide a new way for the research of computer aided diagnosis system.