Characterization of breast masses as benign or malignant at 3.0T MRI with whole-lesion histogram analysis of the apparent diffusion coefficient

Shiteng Suo; Kebei Zhang; Mengqiu Cao; Xinjun Suo; Jia Hua; Xiaochuan Geng; Jie Chen; Zhiguo Zhuang; Xiang Ji; Qing Lu; He Wang; Jianrong Xu

doi:10.1002/jmri.25043

Characterization of breast masses as benign or malignant at 3.0T MRI with whole-lesion histogram analysis of the apparent diffusion coefficient

J Magn Reson Imaging. 2016 Apr;43(4):894-902. doi: 10.1002/jmri.25043. Epub 2015 Sep 7.

Authors

Shiteng Suo¹, Kebei Zhang¹, Mengqiu Cao¹, Xinjun Suo², Jia Hua¹, Xiaochuan Geng¹, Jie Chen¹, Zhiguo Zhuang¹, Xiang Ji³, Qing Lu¹, He Wang⁴, Jianrong Xu¹

Affiliations

¹ Department of Radiology, Ren Ji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China.
² School of Medical Imaging, Tianjin Medical University, Tianjin, China.
³ School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China.
⁴ Philips Research China, Shanghai, China.

PMID: 26343918
DOI: 10.1002/jmri.25043

Abstract

Purpose: To investigate the utility of whole-lesion apparent diffusion coefficient (ADC) histogram analysis in capturing breast lesion heterogeneity and determine which ADC metric may help best differentiate benign from malignant breast mass lesions at 3.0T magnetic resonance imaging (MRI).

Materials and methods: We retrospectively included 101 women with breast mass lesions (benign:malignant = 36:65) who underwent 3.0T diffusion-weighted imaging (DWI) and subsequently had histopathologic confirmation. ADC histogram parameters, including the mean, minimum, maximum, 10th/25th/50th/75th/90th percentile, skewness, kurtosis, and entropy ADCs, were derived for the whole-lesion volume in each patient. Mann-Whitney U-test, univariate and multivariate logistic regression, area under the receiver-operating characteristic curve (Az ), intraclass correlation coefficient (ICC), and Bland-Altman test were used for statistical analysis.

Results: Mean, minimum, maximum, and 10th/25th/50th/75th/90th percentile ADCs were significantly lower (all P < 0.0001), while skewness and entropy ADCs were significantly higher (P < 0.001 and P = 0.001, respectively) in malignant lesions compared with benign ones. The Az values of minimum and 25th percentile ADCs were significantly higher than that of mean ADC (P = 0.0194 and P = 0.0154, respectively) or that of median ADC (P = 0.0300 and P = 0.0401, respectively), indicating that minimum and 25th percentile ADCs may be more accurate for lesion discrimination. Multivariate logistic regression showed that the minimum ADC was the unique independent predictor of breast malignancy. Minimum and 25th percentile ADCs had excellent interobserver agreement (ICC = 0.943 and 0.989, respectively; narrow width of 95% limits of agreement).

Conclusion: These results suggest that whole-lesion ADC histogram analysis may facilitate the differentiation between benign and malignant breast mass lesions.

Keywords: apparent diffusion coefficient; breast mass; diffusion-weighted imaging; heterogeneity; histogram analysis.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Adolescent
Adult
Aged
Algorithms
Breast / diagnostic imaging
Breast / pathology
Breast Neoplasms / diagnostic imaging*
Diffusion Magnetic Resonance Imaging*
Female
Fibroadenoma / diagnostic imaging
Humans
Image Interpretation, Computer-Assisted / methods
Image Processing, Computer-Assisted / methods*
Middle Aged
Motion
Observer Variation
ROC Curve
Regression Analysis
Reproducibility of Results
Retrospective Studies
Statistics, Nonparametric
Young Adult