Semiautomatic estimation of breast density with DM-Scan software

Abstract

Objective: To evaluate the reproducibility of the calculation of breast density with DM-Scan software, which is based on the semiautomatic segmentation of fibroglandular tissue, and to compare it with the reproducibility of estimation by visual inspection.

Material and methods: The study included 655 direct digital mammograms acquired using craniocaudal projections. Three experienced radiologists analyzed the density of the mammograms using DM-Scan, and the inter- and intra-observer agreement between pairs of radiologists for the Boyd and BI-RADS® scales were calculated using the intraclass correlation coefficient. The Kappa index was used to compare the inter- and intra-observer agreements with those obtained previously for visual inspection in the same set of images.

Results: For visual inspection, the mean interobserver agreement was 0,876 (95% CI: 0,873-0,879) on the Boyd scale and 0,823 (95% CI: 0,818-0,829) on the BI-RADS® scale. The mean intraobserver agreement was 0,813 (95% CI: 0,796-0,829) on the Boyd scale and 0,770 (95% CI: 0,742-0,797) on the BI-RADS® scale. For DM-Scan, the mean inter- and intra-observer agreement was 0,92, considerably higher than the agreement for visual inspection.

Conclusion: The semiautomatic calculation of breast density using DM-Scan software is more reliable and reproducible than visual estimation and reduces the subjectivity and variability in determining breast density.

Keywords: Breast cancer; Breast density; Computer-assisted diagnosis; Cáncer de mama; Densidad mamaria; Diagnóstico asistido por ordenador; Digital processing of mammograms; Interobserver variability; Intraobserver variability; Mammography; Mamografía; Procesamiento digital de mamografías; Variabilidad interobservador; Variabilidad intraobservador.

Copyright © 2012 SERAM. Published by Elsevier Espana. All rights reserved.