Improving the Distinction between Benign and Malignant Breast Lesions: The Value of Sonographic Texture Analysis

Abstract

To improve the ability of ultrasound to distinguish benign from malignant breast lesions, we used quantitative analysis of ultrasound image texture. Eight cancers, 22 cysts, 28 fibroadenomata, and 22 fibrocystic nodules were studied. The true nature of each lesion was determined by aspiration (for some cysts) or by open biopsy. Analysis of image texture was performed on digitized video output from the ultrasound scanner using fractal analysis and statistical texture analysis methods. The most useful features were those derived from co-occurrence matrices of the images. Using two features together (contrast of a co-occurrence matrix taken in an oblique direction, and correlation of a co-occurrence matrix taken in the horizontal direction), it was possible to exclude 78% of fibroadenomata, 73% of cysts, and 91% of fibrocystic nodules while maintaining 100% sensitivity for cancer. These findings suggest that ultrasonic image texture analysis is a simple way to markedly reduce the number of benign lesion biopsies without missing additional cancers.

Automatic ultrasound segmentation and morphology based diagnosis of solid breast tumors

“…Texture features are helpful to classify benign and malignant tumors on sonography. The potential of sonographic texture analysis to improve breast tumor diagnosis has already been demonstrated [28][29][30]. However, the texture analysis is system-dependent.…”

The biological correlates of macroscopic breast tumour structure measured using fractal analysis in patients undergoing neoadjuvant chemotherapy

“…The use of measurements of fractal geometry [1] have led to significant progress in understanding complex functional properties, architectural morphology and structural features characterising both normal and abnormal body tissues [2]. These fractal measures, when applied to images of breast cancers, have been able to distinguish between different pathological conditions using ultrasound [3], mammography [4] and, as defined histologically [5]. Rose et al [6] has also demonstrated fractal descriptors of dynamic contrast enhanced MRI (DCE-MRI) and shown that heterogeneity can differentiate between low and high grade malignant tumours.…”

A cost-sensitive extension of AdaBoost with markov random field priors for automated segmentation of breast tumors in ultrasonic images

“…Although numerous algorithms have been proposed [1,2], these algorithms still suffer from errors in segmentation caused by noise, posterior echoes, and lateral shadows. For instance, texture-analysis-based approaches focus on the moments of K-distribution [1] as well as features acquired by co-occurrence matrices [2].…”

“…Although numerous algorithms have been proposed [1,2], these algorithms still suffer from errors in segmentation caused by noise, posterior echoes, and lateral shadows. For instance, texture-analysis-based approaches focus on the moments of K-distribution [1] as well as features acquired by co-occurrence matrices [2]. However, co-occurrence-based features [3] were originally developed for other image processing purposes and were not specified for tumor segmentation in ultrasonic images, which results in low performance.…”