Published in:
01-02-2016 | Breast
Magnetic resonance imaging texture analysis classification of primary breast cancer
Authors:
S. A. Waugh, C. A. Purdie, L. B. Jordan, S. Vinnicombe, R. A. Lerski, P. Martin, A. M. Thompson
Published in:
European Radiology
|
Issue 2/2016
Login to get access
Abstract
Objectives
Patient-tailored treatments for breast cancer are based on histological and immunohistochemical (IHC) subtypes. Magnetic Resonance Imaging (MRI) texture analysis (TA) may be useful in non-invasive lesion subtype classification.
Methods
Women with newly diagnosed primary breast cancer underwent pre-treatment dynamic contrast-enhanced breast MRI. TA was performed using co-occurrence matrix (COM) features, by creating a model on retrospective training data, then prospectively applying to a test set. Analyses were blinded to breast pathology. Subtype classifications were performed using a cross-validated k-nearest-neighbour (k = 3) technique, with accuracy relative to pathology assessed and receiver operator curve (AUROC) calculated. Mann-Whitney U and Kruskal-Wallis tests were used to assess raw entropy feature values.
Results
Histological subtype classifications were similar across training (n = 148 cancers) and test sets (n = 73 lesions) using all COM features (training: 75 %, AUROC = 0.816; test: 72.5 %, AUROC = 0.823). Entropy features were significantly different between lobular and ductal cancers (p < 0.001; Mann-Whitney U). IHC classifications using COM features were also similar for training and test data (training: 57.2 %, AUROC = 0.754; test: 57.0 %, AUROC = 0.750). Hormone receptor positive and negative cancers demonstrated significantly different entropy features. Entropy features alone were unable to create a robust classification model.
Conclusion
Textural differences on contrast-enhanced MR images may reflect underlying lesion subtypes, which merits testing against treatment response.
Key Points
• MR-derived entropy features, representing heterogeneity, provide important information on tissue composition.
• Entropy features can differentiate between histological and immunohistochemical subtypes of breast cancer.
• Differing entropy features between breast cancer subtypes implies differences in lesion heterogeneity.
• Texture analysis of breast cancer potentially provides added information for decision making.