Back to the list
Congress: ECR25
Poster Number: C-13257
Type: Poster: EPOS Radiologist (scientific)
Authorblock: C. Piccolo, M. Sarli, M. Pileri, M. Tommasiello, A. Rofena, V. Guarrasi, P. Soda, B. Beomonte Zobel; Rome/IT
Disclosures:
Claudia Piccolo: Nothing to disclose
Marina Sarli: Nothing to disclose
Matteo Pileri: Nothing to disclose
Manuela Tommasiello: Nothing to disclose
Aurora Rofena: Nothing to disclose
Valerio Guarrasi: Nothing to disclose
Paolo Soda: Nothing to disclose
Bruno Beomonte Zobel: Nothing to disclose
Keywords: Breast, Mammography, Computer Applications-Detection, diagnosis, Cancer
Methods and materials

This retrospective, single-center study (September 2021–June 2023) included 134 women with histologically confirmed breast cancer undergoing contrast-enhanced mammography (CEM) at the Fondazione Policlinico Universitario Campus Bio-Medico, Rome. Inclusion criteria were BI-RADS 4 or 5 findings on conventional imaging, age >18 years, and ability to undergo CEM after informed consent. Exclusion criteria included pregnancy, iodinated contrast allergy, renal failure, and breast prostheses.

CEM was performed using a digital mammography unit with dual-energy imaging following intravenous iodinated contrast injection (Omnipaque 350 mg/mL, 1.5 mL/kg, 2.5 mL/s). Craniocaudal (CC) and mediolateral oblique (MLO) views were acquired bilaterally. If enhancement was observed, an additional image was taken at 8 minutes to assess malignancy probability.

Lesions classified as BI-RADS 4 or 5 were biopsied (core needle or vacuum-assisted), and histological evaluation followed WHO guidelines. Tumor type, grade, receptor status (ER, PgR, HER2), Ki67 index, and nodal involvement were analyzed. HER2 equivocal results (2+) underwent FISH analysis.

Lesions and contours were segmented manually using 3D Slicer, with contour delineations set to 5 mm thickness and 1 mm distance. Radiomic features were extracted using PyRadiomics, focusing on seven feature classes: First Order, Shape (2D), GLCM, GLRLM, GLSZM, NGTDM, and GLDM.

Features were categorized into seven classes:

  • First Order: Statistical measures of pixel intensity distribution (19 features).
  • Shape (2D): Geometric and spatial properties (10 features).
  • GLCM: Texture features based on spatial gray-level relationships (24 features).
  • GLRLM: Measures gray-level run lengths (16 features).
  • GLSZM: Quantifies gray-level zones, capturing heterogeneity (16 features).
  • NGTDM: Local texture variations (5 features).
  • GLDM: Analyzes gray-level dependencies (14 features).

Radiomic features were analyzed for correlations with prognostic factors using univariate and multivariate approaches.
GALLERY