Improved characterization of sub-centimeter enhancing breast masses on MRI with radiomics and machine learning in BRCA mutation carriers

• DOI: 10.1007/s00330-020-06991-7
• Publication Date: 2020-06-27T11:03:14Z
• Authors: Roberto Lo Gullo; Isaac Daimiel; Carolina Rossi Sa...; Almir Bitencourt; Peter Gibbs; Michael J. Fox; Sunitha B. Thakur; Danny F. Martinez; Maxine S. Jochelson; Elizabeth A. Morris; Katja Pinker
• Keywords: Neuroradiology; Breast MRI; BI-RADS; Univariate; Interventional radiology; Univariate analysis; Breast imaging; Breast biopsy

Abstract Objectives To investigate whether radiomics features extracted from MRI of BRCA-positive patients with sub-centimeter breast masses can be coupled machine learning to differentiate benign malignant lesions using model-free parameter maps. Methods In this retrospective study, who had an November 2013 February 2019 that led a biopsy (BI-RADS 4) or imaging follow-up 3) for were included. Two radiologists assessed all independently and in consensus according BI-RADS. Radiomics calculated open-source CERR software. Univariate analysis multivariate modeling performed identify significant clinical factors included model lesions. Results Ninety-six BRCA mutation carriers (mean age at = 45.5 ± 13.5 years) Consensus BI-RADS classification assessment achieved diagnostic accuracy 53.4%, sensitivity 75% (30/40), specificity 42.1% (32/76), PPV 40.5% (30/74), NPV 76.2% (32/42). The combining five parameters (age, lesion location, GLCM-based correlation the pre-contrast phase, first-order coefficient variation 1st post-contrast SZM-based gray level variance phase) 81.5%, 63.2% (24/38), 91.4% (64/70), 80.0% (24/30), 82.1% (64/78). Conclusions improves characterizing as compared qualitative morphological alone carriers. Key Points • help even if are small benign. showed improved accuracy, specificity, PPV, alone.