Evaluating Upstaging in Ductal Carcinoma In Situ Using Preoperative MRI-Based Radiomics

BackgroundAbout 20%-40% of patients diagnosed with ductal carcinoma in situ (DCIS) by core needle biopsy (CNB) will develop invasive cancer at the time of excision. Improving the preoperative diagnosis of DCIS is important for surgical planning.PurposeTo establish an MRI-based radiomics nomogram for preoperatively evaluating the upstaging of DCIS patients and help with risk stratification.Study TypeRetrospective.PopulationA total of 227 patients (50.5 & PLUSMN; 9.7 years; 67 upstaged DCIS) were divided into training (n = 109), internal (n = 47), and external (n = 71) validation cohort.Field Strength/Sequence1.5-T or 3-T, dynamic contrast-enhanced (DCE) imaging, and diffusion-weighted imaging (DWI).AssessmentDCIS lesions were manually segmented using ITK-SNAP software and 1304 radiomic features were extracted from DCE, DWI, and apparent diffusion coef-ficient (ADC) maps, respectively. A radscore was calculated by a random forest algo-rithm based on DCIS upstaging-related radiomic features, which selected by a coarse-to-fine method including interclass correlation coefficient, single-factor anal-ysis, and the least absolute shrinkage and selection operator (LASSO) method. Uni-variate and multivariate logistic regression was used to analyze the independent risk factors, including age, location, lesion size, estrogen receptor (ER) status, and other clinico-pathologic factors. Finally, Mann-Whitney U tests were performed to com-pare the differences in radscore between low/intermediate and high nuclear grade groups for pure DCIS patients.Statistical TestsStudent's t-tests or Mann-Whitney U tests, chi-square-tests, or Fisher's-tests, univariate and multivariate logistic regression analysis, calibration curve, Youden index, the area under the curve (AUC), Delong test, net reclassification improvement (NRI), and integrated discrimination improvement (IDI) analyses.ResultsEight important radiomic features (two from ADC, three from DWI, and three from DCE) were selected for calculating radscore. Clinical model including age and ER was established with AUCs of 0.747 and 0.738 in the internal and external validation cohorts, respectively. A combined model integrating age, estrogen receptor (ER), and radscore were also constructed with AUCs of 0.887 and 0.881. Further subgroup analysis showed that pure DCIS patients with different nuclear grade have significant differences in radscore.Data ConclusionMultisequence MRI radiomics may preoperatively evaluate the upstaging of DCIS and might provide personalized image-based clinical decision support.Evidence Level4.Technical EfficacyStage 2.