Impact of ambiguous breast density assessment on mammographic diagnostic performance and comparison with AI-based quantitative density measurement

BACKGROUND: Dense breast (DB) on mammogram is a risk factor for breast cancer and may hinder the detection of tumor. The U.S. Food and Drug Administration (FDA) issued a final rule requiring mammography reports to include information on breast density. However, the visual assessment of breast density is subjective and may result in considerable inter-observer and intra-observer variability. The objective of this study is to evaluate discrepancies in breast density assessment and to examine the clinical utility of quantitative breast density measurement using artificial intelligence (AI). METHODS: 1,141 mammograms from breast cancer patients were analyzed. Two expert readers independently evaluated the mammographic findings in a double-blind method without referring to clinical information. The classifications of "Scattered fibroglandular densities" and "Heterogeneously dense" were further subdivided based on diagnostic uncertainty into six groups: A, Almost entirely fat (< 25% glandular); B-1, Scattered fibroglandular densities without ambiguity (25-40%); B-2, Scattered fibroglandular densities with ambiguity (41-50%); C-1, Heterogeneously dense with ambiguity (51-60%); C-2, Heterogeneously dense without ambiguity (61-75%); D, Extremely dense (> 75%). Diagnostic performance was evaluated in terms of sensitivity and specificity. Quantitative breast density was also measured using AI-based software developed by EIZO Corporation. RESULTS: Intra-observer agreement was good (κ = 0.76 for Reader I and 0.87 for Reader II). Inter-observer agreement was also good (κ = 0.86). However, complete agreement remained relatively low for the borderline categories (Group B-2: 48% and Group C-1: 42%). Sensitivity and specificity in A, B-1, B-2, and C-1 were high (sensitivity 88.4-94.6%, specificity 85.2-94.4%). However, diagnostic performance decreased in C-2 (sensitivity 83.4% and 83.9%) and D (sensitivity 76.9%,76.9%). AI-based quantitative breast density analysis demonstrated a sensitivity of 81.4% at a PMD cutoff of 62% and 76.8% at 64%, representing a 4.6-percentage-point decrease when increasing the cutoff from 62 to 64%. Raising the PMD cutoff from 50 to 62% resulted in a reduction in the DB ratio from 27.6% to 9.9%. CONCLUSIONS: Visual breast-density assessment was least reproducible in the ambiguous categories between Scattered fibroglandular densities and Heterogeneously dense. Quantitative AI-based PMD assessment may provide a more reproducible measure for examining the relationship between breast density and mammographic sensitivity. Analyses of all cancers and invasive cancers showed an exploratory PMD range that reduced the DB ratio without substantially compromising sensitivity and specificity. However, this finding was not reproduced in the T1-only analysis, and further validation is required before any PMD threshold can be used for dense-breast notification or decisions regarding supplemental screening.

Authors

Institutions

Publication Details

Journal
Breast Cancer
Published
2026-11-16
DOI
https://doi.org/10.1007/s12282-026-01939-z
Primary Topic
Digital Radiography and Breast Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Impact of ambiguous breast density assessment on mammographic diagnostic performance and comparison with AI-based quantitative density measurement

Yuna Fukuma, Naruto Taira, Tsunehisa Nomura
4 links
Breast Cancer
Digital Radiography and Breast Imaging
article

Impact of ambiguous breast density assessment on mammographic diagnostic performance and comparison with AI-based quantitative density measurement

Yuna Fukuma, Naruto Taira, Tsunehisa Nomura
article en

Abstract

BACKGROUND: Dense breast (DB) on mammogram is a risk factor for breast cancer and may hinder the detection of tumor. The U.S. Food and Drug Administration (FDA) issued a final rule requiring mammography reports to include information on breast density. However, the visual assessment of breast density is subjective and may result in considerable inter-observer and intra-observer variability. The objective of this study is to evaluate discrepancies in breast density assessment and to examine the clinical utility of quantitative breast density measurement using artificial intelligence (AI). METHODS: 1,141 mammograms from breast cancer patients were analyzed. Two expert readers independently evaluated the mammographic findings in a double-blind method without referring to clinical information. The classifications of "Scattered fibroglandular densities" and "Heterogeneously dense" were further subdivided based on diagnostic uncertainty into six groups: A, Almost entirely fat (< 25% glandular); B-1, Scattered fibroglandular densities without ambiguity (25-40%); B-2, Scattered fibroglandular densities with ambiguity (41-50%); C-1, Heterogeneously dense with ambiguity (51-60%); C-2, Heterogeneously dense without ambiguity (61-75%); D, Extremely dense (> 75%). Diagnostic performance was evaluated in terms of sensitivity and specificity. Quantitative breast density was also measured using AI-based software developed by EIZO Corporation. RESULTS: Intra-observer agreement was good (κ = 0.76 for Reader I and 0.87 for Reader II). Inter-observer agreement was also good (κ = 0.86). However, complete agreement remained relatively low for the borderline categories (Group B-2: 48% and Group C-1: 42%). Sensitivity and specificity in A, B-1, B-2, and C-1 were high (sensitivity 88.4-94.6%, specificity 85.2-94.4%). However, diagnostic performance decreased in C-2 (sensitivity 83.4% and 83.9%) and D (sensitivity 76.9%,76.9%). AI-based quantitative breast density analysis demonstrated a sensitivity of 81.4% at a PMD cutoff of 62% and 76.8% at 64%, representing a 4.6-percentage-point decrease when increasing the cutoff from 62 to 64%. Raising the PMD cutoff from 50 to 62% resulted in a reduction in the DB ratio from 27.6% to 9.9%. CONCLUSIONS: Visual breast-density assessment was least reproducible in the ambiguous categories between Scattered fibroglandular densities and Heterogeneously dense. Quantitative AI-based PMD assessment may provide a more reproducible measure for examining the relationship between breast density and mammographic sensitivity. Analyses of all cancers and invasive cancers showed an exploratory PMD range that reduced the DB ratio without substantially compromising sensitivity and specificity. However, this finding was not reproduced in the T1-only analysis, and further validation is required before any PMD threshold can be used for dense-breast notification or decisions regarding supplemental screening.

Breast Cancer
Kawasaki Medical School (JP)
Good health and well-being
Digital Radiography and Breast Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon