Click here to


Are you sure ?

Yes, do it No, cancel

Prediction of Benign Or Malignant Breast Masses Using Texture Features From Digital Mammograms by Three Machine Learning Methods

Y Cui12, Y Li3, J Zhu124*, J Dong2*,(1) Department of Radiation Oncology Physics and Technology,Shandong Cancer Hospital affiliated to Shandong University, Jinan 250117, China.(2)Shandong Provincial Key Laboratory of Network based Intelligent Computing, School of Information Science and Engineering, University of Jinan, Jinan 250022, China. (3) Department of Radiology, Shandong Cancer Hospital affiliated to Shandong University,Jinan 250117,China.(4) Shandong Medical Imaging and Radiotherapy Engineering Technology Research Center,Jinan 250117, China


(Tuesday, 7/16/2019) 1:15 PM - 1:45 PM

Room: Exhibit Hall | Forum 8

Purpose: Breast cancer is one of the most common malignant disease for women. Mammography is the preferred method for breast cancer detection. To investigate the feasibility and accuracy of texture features extracted from digital mammograms at predicting benign and malignant breast mass using Radiomics.

Methods: 382 digital mammograms data who was diagnosed as breast masses (benign: 188 malignant: 194) by mammography were enrolled while breast masses were classified as BI-RADS 3, 4, and 5 and confirmed by histopathology. Lesion area was marked with a rectangular frame on the cranio-caudal and mediolateral oblique images at the 5M workstation. The rectangular regions of interest (ROI) was segmented and 4 categories including 455 radiomics features were extracted from each ROI. The dimensionality of the extracted features were reduced by Maximum Relevance Minimum Redundancy (MRMR) algorithm. Post-dimension features were classified using 3 classifiers: Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbor (KNN). 70% and the left 30% of the data was separated into training and testing sets, respectively. The classification effects were evaluated by accuracy, sensitivity and AUC (Area Under Curve).

Results: 30 radiomics features were selected for classification. In testing sets, the classifier of SVM achieved 92.42% accuracy and 90.60% sensitivity and 92.22% AUC; RF achieved an accuracy of 88.91% and a sensitivity of 93.80% and AUC of 89.68%; and the KNN achieved an accuracy of 81.06% and a sensitivity of 79.70% and AUC of 81.37%, respectively.

Conclusion: Through the comparison of radiomics techniques, the SVM classifier achieved better accuracy and sensitivity than RF and KNN. SVM constructs the optimal classification hyperplane to improve the generalization ability of the classifier. The texture features from digital mammograms are expected to predict the benign and malignant breast masses.


Mammography, CAD


IM- Breast x-ray Imaging: CAD

Contact Email