Radiomic Signatures Derived from Diffusion-Weighted Imaging for the Assessment of Breast Cancer Receptor Status and Molecular Subtypes
Doris Leithner,Blanca Bernard-Davila,Danny F. Martinez,Joao V. Horvat,Maxine S. Jochelson,Maria Adele Marino,Daly Avendano,R. Elena Ochoa-Albiztegui,Elizabeth J. Sutton,Elizabeth A. Morris,Sunitha B. Thakur,Katja Pinker
DOI: https://doi.org/10.1007/s11307-019-01383-w
2019-06-17
Molecular Imaging and Biology
Abstract:<h3 class="Heading">Purpose</h3><p class="Para">To compare annotation segmentation approaches and to assess the value of radiomics analysis applied to diffusion-weighted imaging (DWI) for evaluation of breast cancer receptor status and molecular subtyping.</p><h3 class="Heading">Procedures</h3><p class="Para">In this IRB-approved HIPAA-compliant retrospective study, 91 patients with treatment-naïve breast malignancies proven by image-guided breast biopsy, (luminal A, <em class="EmphasisTypeItalic">n</em> = 49; luminal B, <em class="EmphasisTypeItalic">n</em> = 8; human epidermal growth factor receptor 2 [HER2]-enriched, <em class="EmphasisTypeItalic">n</em> = 11; triple negative [TN], <em class="EmphasisTypeItalic">n</em> = 23) underwent multiparametric magnetic resonance imaging (MRI) of the breast at 3 T with dynamic contrast-enhanced MRI, T2-weighted and DW imaging. Lesions were manually segmented on high b-value DW images and segmentation ROIS were propagated to apparent diffusion coefficient (ADC) maps. In addition in a subgroup (<em class="EmphasisTypeItalic">n</em> = 79) where lesions were discernable on ADC maps alone, these were also directly segmented there. To derive radiomics signatures, the following features were extracted and analyzed: first-order histogram (HIS), co-occurrence matrix (COM), run-length matrix (RLM), absolute gradient, autoregressive model (ARM), discrete Haar wavelet transform (WAV), and lesion geometry. Fisher, probability of error and average correlation, and mutual information coefficients were used for feature selection. Linear discriminant analysis followed by k-nearest neighbor classification with leave-one-out cross-validation was applied for pairwise differentiation of receptor status and molecular subtyping. Histopathologic results were considered the gold standard.</p><h3 class="Heading">Results</h3><p class="Para">For lesion that were segmented on DWI and segmentation ROIs were propagated to ADC maps the following classification accuracies > 90% were obtained: luminal B <em class="EmphasisTypeItalic">vs.</em> HER2-enriched, 94.7 % (based on COM features); luminal B <em class="EmphasisTypeItalic">vs.</em> others, 92.3 % (COM, HIS); and HER2-enriched <em class="EmphasisTypeItalic">vs.</em> others, 90.1 % (RLM, COM). For lesions that were segmented directly on ADC maps, better results were achieved yielding the following classification accuracies: luminal B <em class="EmphasisTypeItalic">vs.</em> HER2-enriched, 100 % (COM, WAV); luminal A <em class="EmphasisTypeItalic">vs.</em> luminal B, 91.5 % (COM, WAV); and luminal B <em class="EmphasisTypeItalic">vs.</em> others, 91.1 % (WAV, ARM, COM).</p><h3 class="Heading">Conclusions</h3><p class="Para">Radiomic signatures from DWI with ADC mapping allows evaluation of breast cancer receptor status and molecular subtyping with high diagnostic accuracy. Better classification accuracies were obtained when breast tumor segmentations could be performed on ADC maps.</p>
radiology, nuclear medicine & medical imaging