Abstract Archives of the RSNA, 2007
LL-BR2133-L04
Ultrasound (US) Elasticity Images and Computer-aided Analysis for Classification of Nonpalpable Benign and Malignant Breast Masses
Scientific Posters
Presented on November 28, 2007
Presented as part of LL-BR-L: Breast Imaging
Ji Won Choi MD, Presenter: Nothing to Disclose
Woo Kyung Moon MD, Abstract Co-Author: Nothing to Disclose
Nariya Cho MD, Abstract Co-Author: Nothing to Disclose
Mi Jung Jang MD, Abstract Co-Author: Nothing to Disclose
Kwang Gi Kim PhD, Abstract Co-Author: Nothing to Disclose
To retrospectively evaluate the performance of US elasticity images and computer-aided analysis for classification of biopsy-proved nonpalpable benign and malignant breast tumors
Real-time US elastography was performed in 140 women (mean age,47 yrs;range,24–67 yrs) scheduled for US-guided core biopsy(101 benign,39 malignant tumors). Representative gray-scale and elasticity images of transverse and longitudinal scans were saved as bitmap files. After subtraction of gray-scale images from elasticity images,a region of interest drawn around the margin of mass on gray-scale image was loaded on subtracted color-scale images. The score of each pixel was assigned as from 0 for greatest strain(red) to 255 for no strain(blue). Average, skewness, kurtosis, difference histogram variation(DHV), edge density(ED), and run length were calculated. A neural network was used to classify tumors using these six features. Two breast radiologists provided elasticity score(1-5) by consensus without histologic information. The performance of neural network and radiologists were compared by ROC curve analysis.
The mean values of six elasticity features were different from malignant and benign masses as follows: 235±18 vs 194±38 in average, 264±6 vs 96 ±5 in skewness, 8861±6162 vs 3924±4381 in kurtosis, 7157±4747 vs 109707±64920 in DHV, 1018±9 vs 1004±40 in ED, and 661±133 vs 734±77 in run length(P<.01 in all six features). The sensitivity, specificity, and PPV and NPV were 91%,74%,58%,and 95% for neural network based on all six elasticity features and 97%,40%,38%,and 97% at cutoff score between 2 and 3 and 54%,91%,70%,and 84% at cutoff score between 3 and 4 for radiologists. The Az value was 0.89 for neural network and 0.81 for radiologists and the difference was significant(P<.02).
Computer-aided analysis of US elasticity images showed better performance than radiologists for classification of nonpalpable benign and malignant breast tumors.
Computer-aided analysis of US elasticity images can be used an objective method to evaluate tissue strain and can aid in classification of nonpalpable benign and malignant breast tumors.
Choi, J,
Moon, W,
Cho, N,
Jang, M,
Kim, K,
Ultrasound (US) Elasticity Images and Computer-aided Analysis for Classification of Nonpalpable Benign and Malignant Breast Masses. Radiological Society of North America 2007 Scientific Assembly and Annual Meeting, November 25 - November 30, 2007 ,Chicago IL.
http://archive.rsna.org/2007/5014985.html