Architectural distortion recognition based on a subclass technique and the sparse representation classifier

Architectural distortion is the third most common sign of breast cancer in mammograms. The accurate recognition is important for computer aided diagnosis of breast cancer. However, due to the subtle symptom and complex structures in the mammogram images, it is difficult to recognize whether a region of interest (ROI) is truly an architectural distortion. In this paper, we proposed a new method for architectural distortion recognition. In the proposed method, several texture features are extracted for each region of interest, including features from GLCM matrix, spiculated related features, entropy features, etc. Feature selection is obtained by a sub-classes clustering based multi-task learning method (SMTL), which can utilize the discriminative label information and reflect the multi-clustering characteristic of the data samples. Finally, the powerful sparse representation based classifier is used for the classification of AD or non-AD. The proposed method has been tested on DDSM dataset and compared with several other methods, the experimental results showed the effectiveness of the proposed method.