DCE‐MRI texture analysis with tumor subregion partitioning for predicting Ki‐67 status of estrogen receptor‐positive breast cancers

Abstract

BACKGROUND: Breast tumor heterogeneity is related to risk factors that lead to worse prognosis, yet such heterogeneity has not been well studied. PURPOSE: To predict the Ki-67 status of estrogen receptor (ER)-positive breast cancer patients via analysis of tumor heterogeneity with subgroup identification based on patterns of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). STUDY TYPE: Retrospective study. POPULATION: Seventy-seven breast cancer patients with ER-positive breast cancer were investigated, of whom 51 had low Ki-67 expression. FIELD STRENGTH/SEQUENCE: -weighted 3.0T DCE-MR images. ASSESSMENT: Each tumor was partitioned into multiple subregions using three methods based on patterns of dynamic enhancement: 1) time to peak (TTP), 2) peak enhancement rate (PER), and 3) kinetic pattern clustering (KPC). In each tumor subregion, 18 texture features were computed. STATISTICAL TESTING: Univariate and multivariate logistic regression analyses were performed using a leave-one-out-based cross-validation (LOOCV) method. The partitioning results were compared with the same feature extraction methods across the whole tumor. RESULTS: In the univariate analysis, the best-performing feature was the texture statistic of sum variance in the tumor subregion with early TTP for differentiating between patients with high and low Ki-67 expression (area under the receiver operating characteristic curves, AUC = 0.748). Multivariate analysis showed that features from the tumor subregion associated with early TTP yielded the highest performance (AUC = 0.807) among the subregions for predicting the Ki-67 status. Among all regions, the tumor area with high PER at a precontrast MR image achieved the highest performance (AUC = 0.722), while the subregion that exhibited the highest overall enhancement rate based on KPC had an AUC of 0.731. These three models based on intratumoral texture analysis significantly (P < 0.01) outperformed the model using features from the whole tumor (AUC = 0.59). DATA CONCLUSION: Texture analysis of intratumoral heterogeneity has the potential to serve as a valuable clinical marker to enhance the prediction of breast cancer prognosis. LEVEL OF EVIDENCE: 4 Technical Efficacy: Stage 3 J. Magn. Reson. Imaging 2017.