Document Type : Original Paper
Authors
1
Pharmaceutical Research Center, Mashhad University of Medical Sciences, Mashhad, Iran
2
Department of Radiology and Biomedical Imaging, Yale School of Medicine, PO Box 208048, New Haven, CT, 06520-8048, USA.
3
Department of Medical Physics, School of Medicine, Iran University of Medical Sciences, Tehran, Iran
4
Department of Radiation Physics, Division of Radiation Oncology, University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA
Abstract
Purpose: Breast cancer is the most common cancer among women worldwide. Early detection of breast cancer reduces mortality and morbidity. Acquiring or identifying valuable information in the form of rules is the key to accurate diagnosis and differentiation between benign and malignant breast cancer. Our goal is to find hidden but beneficial knowledge in the form of rules from datasets. In this paper, we use association rule mining algorithms to obtain information in the form of data rules for differential diagnosis between benign and malignant breast cancer based on radiomics features extracted from mammography images by clinical data. Materials and Methods: In this study, 703 patients with both benign and malignant tumors were selected from the CBIS-DDSM database. The embedded method was employed to select the radiomic features of the image texture and uncover the hidden patterns of data through the Apriori algorithm. Results: The association rules were generated from separated rules for benign and malignancy classes. The important features of the benign class include mass margins, horizontal long-run emphasis, 135drfraction, WavEnHLs3, vertical short-run emphasis, Teta1, 45dgr run-length with no uniformity, Teta2, and differential entropy2. However, the important features of the malignant class include assessment, correlation4, Teta1, WavEnLHs3, contrast5, vertical short-run emphasis, and differential entrpy2. Conclusion: It can be concluded that the proposed method succeeded in determining what features are proprietary attributes for tumor prediction.
Keywords
Main Subjects
)