Open AccessA radiomics study of textural features using magnetic resonance imaging for classification of breast cancer subtypes
Author(s) -
Zhuoyue Tang,
Lloyd Kuan Rui Tan,
Bernard Ng,
Kartini Rahmat,
Marlina Tanty Ramli,
Kenta Ninomiya,
Jeannie Hsiu Ding Wong
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1497/1/012015
Subject(s) - breast cancer , magnetic resonance imaging , mammography , radiomics , medicine , breast mri , logistic regression , radiology , pattern recognition (psychology) , cancer , artificial intelligence , computer science
Breast cancer is usually screened using mammography and biopsy is used to confirm diagnosis. Recent radiomics approaches suggest predictive associations between images and medical outcome. This study aims to classify breast cancer subtypes using textural features derived from magnetic resonance imaging (MRI). Thirty-two lesions with histologic results that were definite were studied. A total of 174 textural features were extracted from four MRI sequences (Axial STIR, dynamic contrast enhance (DCE) Phase 2, dynamic contrast enhance (DCE) subtracted Phase 2 and T1-weighted), and analysed using t-test, Kruskal-Wallis and principal component analysis (PCA). Evaluation was done using multinomial logistic regression and leave-one-out-cross-validation (LOOCV) methods. We found 14 texture features that consistently showed significant difference between malignant and normal breast tissues across all MRI sequences. Four textural features were useful in histological status with t -test accuracy of 71.4% and PCA accuracy of 64.3%. In hormonal receptor status, only five textural features were useful. The accuracies were also found to be poorer with 46.4% accuracy based on Kruskal-Wallis method and 46.4% accuracy using PCA method. As this is a preliminary study, the analysis should be extended to a larger sample size to accurately determine the possibility of clinical diagnosis.