Testing realism of software breast phantoms: texture analysis of synthetic mammograms
Lists
Only you see this
Abstract
Software breast phantoms have been developed for pre-clinical validation of breast imaging systems. Realism is of great importance for the acceptance and the range of applications of breast phantoms. In this paper we have assessed the phantom realism based upon the analysis of mammographic texture properties. Texture analysis is of interest since it reflects the spatial tissue distribution, which is known to correlate with breast cancer risk. We compared texture properties of synthetic mammograms generated using software breast phantoms with clinical data. A total of 133 phantom images were synthesized using software phantoms developed at the University of Pennsylvania. The phantoms were designed using two different anatomy simulation methods: an octree-based recursive partitioning method and a region growing method. The synthetic images were generated assuming a clinically used acquisition geometry and mono-energetic x-ray beam with no scatter. The clinical data included 60 anonymized mammograms selected retrospectively from screening cases at the University of Pennsylvania. The same postprocessing was applied to clinical and phantom images. The texture analysis was performed using fully automated software which extracts a battery of features from analyzed images. The histograms of texture properties extracted from phantom images were compared with those from clinical mammograms, separately for the two anatomy simulation methods. The histogram agreement was quantified using symmetrized Kulback-Leibler divergence. We observed good agreement for most of the analyzed 25 features. In more than a half of the features, the octree-based simulation method yielded better agreement with clinical data as compared with the region growing method.
BibTeX
@inproceedings{bakic2013testing,
author = {Predrag R. Bakić and Brad M. Keller and Yuanjie Zheng and Yan Wang and James C. Gee and Despina Kontos and Andrew D. A. Maidment},
title = {{Testing realism of software breast phantoms: texture analysis of synthetic mammograms}},
booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
year = {2013},
doi = {10.1117/12.2008173},
}Loading related content...