Breast lesion characterization using whole-lesion histogram analysis with stretched-exponential diffusion model

Chunling Liu, Kun Wang, Xiaodan Li, Jine Zhang, Jie Ding, Karl Spuhler, Timothy Duong, Changhong Liang, Chuan Huang

Research output: Contribution to journalArticlepeer-review

22 Scopus citations


Background: Diffusion-weighted imaging (DWI) has been studied in breast imaging and can provide more information about diffusion, perfusion and other physiological interests than standard pulse sequences. The stretched-exponential model has previously been shown to be more reliable than conventional DWI techniques, but different diagnostic sensitivities were found from study to study. Purpose: This work investigated the characteristics of whole-lesion histogram parameters derived from the stretched-exponential diffusion model for benign and malignant breast lesions, compared them with conventional apparent diffusion coefficient (ADC), and further determined which histogram metrics can be best used to differentiate malignant from benign lesions. Study Type: This was a prospective study. Population: Seventy females were included in the study. Field Strength/Sequence: Multi-b value DWI was performed on a 1.5T scanner. Assessment: Histogram parameters of whole lesions for distributed diffusion coefficient (DDC), heterogeneity index (α), and ADC were calculated by two radiologists and compared among benign lesions, ductal carcinoma in situ (DCIS), and invasive carcinoma confirmed by pathology. Statistical Tests: Nonparametric tests were performed for comparisons among invasive carcinoma, DCIS, and benign lesions. Comparisons of receiver operating characteristic (ROC) curves were performed to show the ability to discriminate malignant from benign lesions. Results: The majority of histogram parameters (mean/min/max, skewness/kurtosis, 10–90 th percentile values) from DDC, α, and ADC were significantly different among invasive carcinoma, DCIS, and benign lesions. DDC 10% (area under curve [AUC] = 0.931), ADC 10% (AUC = 0.893), and α mean (AUC = 0.787) were found to be the best metrics in differentiating benign from malignant tumors among all histogram parameters derived from ADC and α, respectively. The combination of DDC 10% and α mean , using logistic regression, yielded the highest sensitivity (90.2%) and specificity (95.5%). Data Conclusion: DDC 10% and α mean derived from the stretched-exponential model provides more information and better diagnostic performance in differentiating malignancy from benign lesions than ADC parameters derived from a monoexponential model. Level of Evidence: 2. Technical Efficacy: Stage 2. J. Magn. Reson. Imaging 2018;47:1701–1710.

Original languageEnglish (US)
Pages (from-to)1701-1710
Number of pages10
JournalJournal of Magnetic Resonance Imaging
Issue number6
StatePublished - Jun 2018
Externally publishedYes


  • apparent diffusion coefficient
  • breast neoplasm
  • diffusion-weighted magnetic resonance imaging
  • distributed diffusion coefficient
  • histogram analysis
  • stretched-exponential diffusion

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging


Dive into the research topics of 'Breast lesion characterization using whole-lesion histogram analysis with stretched-exponential diffusion model'. Together they form a unique fingerprint.

Cite this