Characterisation of mammographic masses using a new Spiculated Mass Descriptor in computer aided diagnosis systems
Abstract
Automatic breast mass recognition as malignant or benign is addressed in order to assist radiologists to make decision. Several researches was based on roughness measures to characterise breast mass boundaries. However, used features generally do not consider specific characteristics of mass spiculations. In this context, we propose a new shape descriptor allowing to characterise simultaneously the number, the depth and the width of spicules. Consequently, the proposed Spiculated Mass Descriptor (SMD) allows to distinguish between ambiguous cases such as lobulated benign masses and microlobulated malignant masses which are hard to classify using common shape analysis methods. In addition, it ensures invariance to geometrical transformations which preserve a robust descriptor insensitive to shifts, orientations and scalings. SMD efficiency is evaluated on the known Digital Database for Screening Mammography (DDSM) using the area under the Receiver Operating Characteristic (ROC) curve analysis. Experimental results show that the new descriptor outperforms several shape features and provides satisfying classification results of benign and malignant masses.