Malignant versus benign tumor classification based on ultrasonic B-scan images of the breast

The Power-law Shot Noise (PLSN) model has been recently proposed for modeling the ultrasound RF echo. Its power-law exponent parameter, which was linked to tissue attenuation, has been effectively used in characterizing normal and abnormal tissue. The K-distribution model has also been proposed in the past for modeling the echo envelope, and functions of its parameters, linked to scatterer density, have been used to discriminate between normal and abnormal tissue. However, neither model produced satisfactory stand-alone features for differentiating between benign and malignant tumors. Observing PLSN and K parameters are linked to different tissue properties, we here propose to use vectors consisting of combinations of these parameters as signatures for malignant versus benign tissue characterization. We test the performance of the proposed features using receiver operating characteristic analysis of 100 clinical images of the breast.