Breast tissue characterization based on modeling of ultrasonic echoes using the power-law shot noise model

Abstract

Ultrasound is a safe, inexpensive and widely available tissue imaging modality, but its use in breast cancer detection is rather limited. Currently there is a significant amount of research devoted to extracting objective features from the ultrasound return signal, to be used in automatic tissue characterization. In this paper we investigate the parameters of the power-law shot noise (PLSN) model that was proposed in [IEEE Trans. UFFC 48 (2001) 953] as potential tissue characterization features. The PLSN model parameters are estimated from a database of 100 clinical ultrasound images of the breast, taken from 25 patients, and receiver operating characteristic (ROC) analysis is subsequently applied to quantify their ability to differentiate between tumorous and non-tumorous tissue and also between malignant and benign tumors. The obtained results indicate a maximum ROC area of 97% for the tumorous versus non-tumorous decision, while a maximum ROC area of 81% is obtained for the benign versus malignant decision. The latter result can be further improved by using the PLSN model parameters along with parameters of the K-distribution model, the combination yielding a maximum ROC area of 88.9%.