Related Experiment Video
Updated: Sep 23, 2025
15:48
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
22.6K
Prediction of Breast Cancer Histological Outcome by Radiomics and Artificial Intelligence Analysis in
Antonella Petrillo1, Roberta Fusco2, Elio Di Bernardo2
1Radiology Division, Istituto Nazionale Tumori-IRCCS-Fondazione G. Pascale, 80131 Naples, Italy.
Cancers
|May 14, 2022
Summary
Radiomics features from contrast-enhanced mammography accurately differentiate malignant from benign breast lesions and predict tumor characteristics. This analysis aids in diagnosing breast cancer subtypes and grading with high accuracy.
Area of Science:
- Radiology
- Medical Imaging
- Oncology
Background:
- Accurate breast cancer diagnosis and characterization are crucial for effective treatment planning.
- Contrast-Enhanced Mammography (CEM) offers improved lesion visualization.
- Radiomics, the extraction of quantitative features from medical images, shows promise in non-invasively characterizing breast lesions.
Purpose of the Study:
- To evaluate radiomics features for differentiating malignant versus benign breast lesions.
- To predict tumor grading (low vs. moderate/high).
- To identify hormone receptor (HR) and human epidermal growth factor receptor 2 (HER2) status.
Main Methods:
- Retrospective study of 182 patients with known breast lesions who underwent CEM.
- Pathology served as the reference standard (118 malignant, 64 benign).
- 837 textural radiomics metrics extracted from craniocaudal (CC) and mediolateral oblique (MLO) views; machine learning algorithms applied.
Main Results:
- Univariate analysis: original_gldm_DependenceNonUniformity (CC view) achieved 88.98% accuracy for malignant/benign classification.
- Multivariate analysis: CC view images with two or more features reached 95.83% accuracy for malignant/benign classification.
- MLO view images: classification tree algorithm predicted grading with 91.67% accuracy; HR prediction achieved 81.65% accuracy; HER2 prediction reached 89.29% accuracy.
Conclusions:
- Radiomics analysis of CEM images can accurately identify malignant breast lesions.
- Histological outcomes and molecular subtypes, particularly HR-positive tumors, can be differentiated with satisfactory accuracy.
- Both univariate and multivariate radiomics analyses are valuable for breast cancer characterization.
