Breast MRI radiomics: comparison of computer- and human-extracted imaging phenotypes

Abstract

BackgroundIn this study, we sought to investigate if computer-extracted magnetic resonance imaging (MRI) phenotypes of breast cancer could replicate human-extracted size and Breast Imaging-Reporting and Data System (BI-RADS) imaging phenotypes using MRI data from The Cancer Genome Atlas (TCGA) project of the National Cancer Institute.MethodsOur retrospective interpretation study involved analysis of Health Insurance Portability and Accountability Act-compliant breast MRI data from The Cancer Imaging Archive, an open-source database from the TCGA project. This study was exempt from institutional review board approval at Memorial Sloan Kettering Cancer Center and the need for informed consent was waived. Ninety-one pre-operative breast MRIs with verified invasive breast cancers were analysed. Three fellowship-trained breast radiologists evaluated the index cancer in each case according to size and the BI-RADS lexicon for shape, margin, and enhancement (human-extracted image phenotypes [HEIP]). Human inter-observer agreement was analysed by the intra-class correlation coefficient (ICC) for size and Krippendorff’s α for other measurements. Quantitative MRI radiomics of computerised three-dimensional segmentations of each cancer generated computer-extracted image phenotypes (CEIP). Spearman’s rank correlation coefficients were used to compare HEIP and CEIP.ResultsInter-observer agreement for HEIP varied, with the highest agreement seen for size (ICC 0.679) and shape (ICC 0.527). The computer-extracted maximum linear size replicated the human measurement with p < 10−12. CEIP of shape, specifically sphericity and irregularity, replicated HEIP with both p values < 0.001. CEIP did not demonstrate agreement with HEIP of tumour margin or internal enhancement.ConclusionsQuantitative radiomics of breast cancer may replicate human-extracted tumour size and BI-RADS imaging phenotypes, thus enabling precision medicine.

Artificial intelligence in the interpretation of breast cancer on MRI

“…This association with breast cancer risk suggests a possible mechanism of correlating underlying biological processes with resultant imaging phenotypes. Deep learning is also being examined for use in assessing breast density and in describing the parenchymal patterns of density on mammograms (Fig. a–c).…”

Background parenchymal enhancement on breast MRI: A comprehensive review

“…In applications for breast cancer risk assessment, radiomic features have emerged as surrogate measures for breast parenchymal patterns in mammographic and tomographic images . With the expanding use of DCE‐MRI for breast cancer screening, recent studies have suggested associations between radiomic characterizations of BPE and breast cancer risk . Specifically, in each study radiomic features characterizing the morphology, texture, and kinetics of BPE, as well as features quantifying breast density, demonstrated a high correlation with qualitative phenotypic characteristics determined by a radiologist, emphasizing the complementary value of radiomics for breast cancer risk assessment.…”

Application of Radiomics and Decision Support Systems for Breast MR Differential Diagnosis

“…Obviously CAD can be considered part of radiomics, but in contrast to CAD's simplicity and ability for answering only simple clinical questions, radiomic analysis considers more complex computational processes aiding decision support, by utilizing a plethora of quantitative imaging features—potential imaging biomarkers, extracted from digital images [ 26 , 28 ]. Furthermore, the correlation of these large-scale radiological phenotypic characteristics with the rich breast histopathological data available, e.g., the expression statuses of estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor 2 receptor (HER2), and triple negative (lack of expression of ER, PR, and HER2), facilitates their strong association with molecular subtypes, which eventually results in the generation of pathology prognostic and predictive models [ 4 , 26 , 27 , 29 ].…”

“…Higher order statistics derived features, referred to as textural features, have been widely utilized in breast tumor DCE-MRI parametric maps in the past, for improving characterization of breast lesions and their response to treatment [ 43 – 47 ]. Second-order histograms such as gray-level co-occurrence matrices (GLCMs) [ 28 , 48 ] and gray-level run-length matrices (GLRLMs) [ 29 , 49 ] characterize spatial relationships between pixel intensities in different 2D or 3D directions and thus are robust in quantifying tumor structural properties and various patterns of heterogeneity. In particular, GLCM analysis of DCE MR data has been proved to be robust in differentiating between benign and malignant breast lesions [ 50 ].…”