Mammary tumor classification using envelope statistics from ultrasound backscatter and the homodyned K distribution.

Abstract

Envelope statistics from ultrasound backscatter based on the homodyned K distribution were used to characterize organizational patterns of tissue microstructure in three rodent models of breast cancer: a mouse mammary carcinoma, a mouse mammary sarcoma, and a rat fibroadenoma. The homodyned K distribution was used to model the amplitude of the envelope of backscattered ultrasound yielding two parameters to characterize tissues: one related to the effective scatterer density and the other related to periodicity of scatterer locations. Ten tumors of each of the three kinds of tumors were scanned. Envelope statistics model parameters were estimated for regions of interest measuring 1 × 1 mm in each of the tumor images. These estimates were then averaged together for each tumor. A linear combination of the two estimated parameters was derived for each tumor in order to obtain a lumped parameter for discriminating between the different kinds of tumors. This linear combination yielded the following (mean±SD): carcinomas 9.58±1.17, sarcomas 12.01±0.78, and fibroadenomas 11.66±1.15. These estimates were statistically significant in distinguishing carcinomas from sarcomas (p<0.05) and carcinomas from fibroadenomas (p<0.05), but a statistically significant difference was not observed between the sarcomas and fibroadenomas (p=0.44). [Work supported by NIH CA111289.]