A HYBRID METHOD FOR MAMMOGRAPHY MASS DETECTION BASED ON WAVELET TRANSFORM

Abstract

Introduction: Breast cancer is a leading cause of death among females throughout the world. Currently, radiologists are able to detect only 75% of breast cancer cases. Making use of computer-aided design (CAD) can play an important role in helping radiologists perform more accurate diagnoses. Material and Methods: Using our hybrid method, the background and the pectoral muscle were removed from mammography images and image contrast was enhanced using an adaptive density weighted method. First, suspected regions were extracted based on mathematical morphology and adaptive thresholding approaches. Then, in order to reduce the false positives in the suspected regions obtained in the first stage, the corresponding features were extracted using a wavelet transform, followed by the application of a support vector machine to detect masses. Results: A Mammographic Image Analysis Society (MIAS) database was used to evaluate the performance of the algorithm. The sensitivity of 81% and specificity of 84% were achieved in detecting masses. Improvement of sensitivity and specificity with our proposed hybrid algorithm was demonstrated by subjective expert-based and objective ROC-based techniques in comparison with the currently acceptable method by Masotti. Disscusion and Conclusion: In this paper, a hybrid method of pixel-based and region-based mass detection approaches is proposed to increase the specificity of the results. The accessory stage (using wavelet features) increased the sensitivity by 30%. It can be concluded that the proposed algorithm is an efficient and robust method for different types of mass detection in low-quality mammography images.