Radiomic and Artificial Intelligence Analysis with Textural Metrics, Morphological and Dynamic Perfusion Features Extracted by Dynamic Contrast-Enhanced Magnetic Resonance Imaging in the Classification of Breast Lesions

Purpose: The aim of the study was to estimate the diagnostic accuracy of textural, morphological and dynamic features, extracted by dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) images, by carrying out univariate and multivariate statistical analyses including artificial intelligence approaches. Methods: In total, 85 patients with known breast lesion were enrolled in this retrospective study according to regulations issued by the local Institutional Review Board. All patients underwent DCE-MRI examination. The reference standard was pathology from a surgical specimen for malignant lesions and pathology from a surgical specimen or fine needle aspiration cytology, core or Tru-Cut needle biopsy for benign lesions. In total, 91 samples of 85 patients were analyzed. Furthermore, 48 textural metrics, 15 morphological and 81 dynamic parameters were extracted by manually segmenting regions of interest. Statistical analyses including univariate and multivariate approaches were performed: non-parametric Wilcoxon–Mann–Whitney test; receiver operating characteristic (ROC), linear classifier (LDA), decision tree (DT), k-nearest neighbors (KNN), and support vector machine (SVM) were utilized. A balancing approach and feature selection methods were used. Results: The univariate analysis showed low accuracy and area under the curve (AUC) for all considered features. Instead, in the multivariate textural analysis, the best performance (accuracy (ACC) = 0.78; AUC = 0.78) was reached with all 48 metrics and an LDA trained with balanced data. The best performance (ACC = 0.75; AUC = 0.80) using morphological features was reached with an SVM trained with 10-fold cross-variation (CV) and balanced data (with adaptive synthetic (ADASYN) function) and a subset of five robust morphological features (circularity, rectangularity, sphericity, gleaning and surface). The best performance (ACC = 0.82; AUC = 0.83) using dynamic features was reached with a trained SVM and balanced data (with ADASYN function). Conclusion: Multivariate analyses using pattern recognition approaches, including all morphological, textural and dynamic features, optimized by adaptive synthetic sampling and feature selection operations obtained the best results and showed the best performance in the discrimination of benign and malignant lesions.