Radiomic features analysis by digital breast tomosynthesis and contrast-enhanced dual-energy mammography to detect malignant breast lesions. (August 2019)

• Record Type: Journal Article
• Authors: Fusco, Roberta; Vallone, Paolo; Filice, Salvatore; Granata, Vincenza; Petrosino, Teresa; Rubulotta, Maria Rosaria; Setola, Sergio Venanzio; Maio, Francesca; Raiano, Concetta; Raiano, Nicola; Siani, Claudio; Di Bonito, Maurizio; Sansone, Mario; Botti, Gerardo; Petrillo, Antonella
• Is Part Of: Biomedical signal processing and control. Volume 53(2019)
• Publication Date: 2019-08
• Subjects: Breast cancer -- Mammography -- CEDM -- Tomosynthesis; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101568
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Lesion texture and morphological features can capture often-missed information regarding the characteristics of a tumor malignancy. Lesion texture and morphological features can provide details that have prognostic or diagnostic value. Radiomic texture features by CEDM and morphological features by DBT showed a potential in differentiating malignant to benign lesions. Abstract: Purpose: To detect malignant breast lesions using radiomic morphological features from Digital Breast Tomosynthesis (DBT) and radiomic textural features from Contrast-enhanced Dual-Energy Digital Mammography (CEDM). Methods: In a 8-month period, we enrolled 72 consecutive patients with breast lesions; their age ranging from 26 to 72 years (mean, 52.2; standard deviation 11.1). Ninety-three breast lesions subjected to CEDM and DBT in cranio caudal (CC) and mediolateral oblique (MLO) view were included: 36 histopathologically proven benign lesions and 59 histopathologically proven malignant lesions were analyzed. We considered a feature set including 23 textural features calculated on CEDM and 14 morphological features extracted by DBT. Non-parametric statistics, receiver operating characteristic with area under curve (AUC), Spearman correlation coefficient and Bonferroni correction were applied. Results: At univariate analysis, the area under ROC was obtained by the best textural feature, the contrast with a value of 0.78. To differentiate malignant lesions with different grading only one textural feature had significant results: median absolute deviation (MAD) (p < 0.01 at Kruskal Wallis test). As a morphological feature by DBT, at univariate analysis, the best area under ROC was obtained by angularity with a value of 0.74. Using morphological parameters there were no statistically significant differences among malignant lesions with different grading. At bivariate analysis using couple combinations of features did not increase the accuracy with respect to single feature. The cross validated decision tree considering the best textural feature (the contrast) and the best morphological feature (the angularity) showed an area under ROC of 0.90, an accuracy of 87.1%, a true positive rate of 84% and a false positive rate of 12%. Considering all texture and morphological metrics with pattern recognition approach was not obtained an increase of diagnostic accuracy. Conclusions: Radiomic textural features from CEDM and radiomic morphological features from DBT have shown a good power to differentiate malignant to benign lesions. A decision tree considering the contrast as textural parameter and the angularity as morphological metric reached the best results (87% of accuracy).