Unencrypted login | home

Program Information

Bayesian Classifier for Differentiating Fibroadenoma and Carcinoma in Breast


J Zagzebski

J Zagzebski*, H Gerges-Nasief, I Rosado-Mendez, S Kohn, T Hall, University of WI-Madison, Madison, WI

SU-D-134-7 Sunday 2:05PM - 3:00PM Room: 134

Purpose:Quantitative ultrasound (QUS) provides acoustic parameters that are closely related to tissue microstructure and sound wave propagation. This study evaluated a Bayesian classifier applied to breast QUS data to aid in differentiating benign from malignant masses. This presentation outlines methodology for deriving these quantitative parameters and reports preliminary results using the classifier on a small data set of human tumors.

Methods:This IRB-approved, HIPAA compliant study recruited 35 subjects who were scheduled for core biopsy of a suspicious breast mass. A Siemens S2000 scanner equipped with linear array transducers was used to acquire radiofrequency echo data at beam-steered angles from -10 to 10 degrees from the normal to the transducer surface in both longitudinal and transverse planes. Echo signals also were obtained from a reference phantom using the same settings. Parameters included acoustic attenuation (ATT), backscatter coefficients (BSC), effective scatterer diameter (ESD), and a scatterer size heterogeneity index(HI). ATT and BSC within the tumor were estimated using a reference phantom method, while theESD was estimated from the BSC using a Gaussian form factor. HI was taken to be the standard deviation among the ESD estimates. A Bayesian classifier was designed to differentiate between Fibroadenoma and carcinoma. The data were divided into training (18 patients) and testing (17 patients) sets. Classification was performed using the minimum distance to the centroid of the training class and the results were compared with the biopsy results.

Results:The best performance with a parameter pair (ATT, ESD) was about 80% which increased to about 100% with three parameters (ATT, ESD, HI). However, more verification studies are needed to include a larger data set and more specific tumor categories.

Conclusion:QUS with a Bayesian classifier is a promising noninvasive tool to characterize breast tumors.

Funding Support, Disclosures, and Conflict of Interest: This work was supported, in part, by NIH (grants R01CA111289, R21HD061896, R21HD063031, and R01HD072077) and the Consejo Nacional de Ciencia y Tecnologia of Mexico (Reg. 206414).

Contact Email: