Do Saliency Maps Fail Dense Breasts? A Chance-Normalized, Confounder-Aware Audit of Explanation Fairness in Mammography

Ahmed Jaheen

cs.CVNov 4, 2026

Do Saliency Maps Fail Dense Breasts? A Chance-Normalized, Confounder-Aware Audit of Explanation Fairness in Mammography

Authors: Ahmed Jaheen

Organizations: New York University, New York, NY, USA

Abstract

Saliency maps, offered as evidence that mammography classifiers use the cancer, are untested in dense breasts, which can hide cancers. We tested whether they locate cancers worse there, and why. On 1,345 malignant CBIS-DDSM views, we scored three CNNs' Grad-CAM, Integrated Gradients and occlusion maps by whether their peak hits the lesion and a chance-normalized lesion-to-background saliency ratio (bits; +1 doubles it). Unadjusted gaps were large: Grad-CAM peaks hit 44% of dense vs. 58% of non-dense lesions. Case mix, chiefly more calcifications in dense breasts, explained about half or more of each gap. Occlusion kept a significant average of -0.56 bit gap. Controls also implicate the classifiers: unadjusted gaps exceeded an image-only conspicuity map's, and models relied less on dense-breast lesions. We recommend reporting density gaps before and after case-mix adjustment, per classifier, with conspicuity and lesion-reliance controls.

Figures & tables

Figure 1: Study overview. Cyan: lesion (solid) and 5-mm zone (dashed). Images: the two mass views of Fig. 4 , selected by its rule; steps 3 and 6 show the dense view (orange frame), with network maps from ResNet-50. M0 to M3 add factors one at a time and re-estimate the gap (Fig. 3 ).

Figure 2: (a) One view (mass, density 2) with two illustrative map peaks. Zone/breast is the pointing-game chance level $\pi_{\mathrm{PG}}$ ; lesion/breast is the energy-lift chance level $\pi$ . (b) $\pi$ by BI-RADS density. Calcification outlines enclose the whole cluster, so their $\pi$ is larger. $\pi$ varies more within groups than between densities, so each view is scored against its own chance level.

Energy lift $\lambda_{E}$ (bit)
	Crude gap	Adjusted gap (M1)
Grad-CAM	$-$ 0.86 [ $-$ 1.23, $-$ 0.51]	$-$ 0.15 [ $-$ 0.53, 0.23]
IG	$-$ 0.62 [ $-$ 0.80, $-$ 0.45]	$+$ 0.10 [ $-$ 0.14, 0.35]
Occlusion	$-$ 1.27 [ $-$ 1.72, $-$ 0.84]	$-$ 0.56 [ $-$ 1.00, $-$ 0.12]
Pointing game (hits per 100 cancers)
Grad-CAM	$-$ 14.1 [ $-$ 19.3, $-$ 9.2]	$-$ 7.6 [ $-$ 12.4, $-$ 2.0]

Table 1: Density gaps, dense minus non-dense [95% CI]; below 0: worse in dense breasts. Crude: group-mean difference. Adjusted: M1 with shared case-mix slopes. LSR: lesion-specific reliance (1,276 views with a comparison region); adjusted LSR exploratory.

Figure 3: Energy-lift gap in each model; the rows under the axis show the factors each model adds. Bars: 95% CIs; case-mix slopes shared across methods. With method-specific slopes (exploratory), no M1 gap is significant. Grey band:

\pm

0.5-bit equivalence margin, tested at M1 with 90% CIs. M0 is the crude gap as the regression estimates it, so it differs slightly from Table 1 .

Figure 4: Rule-selected examples. For each density and lesion-type group ( $n$ views; Calc.: calcification): the view whose ResNet-50 Grad-CAM $\lambda_{E}$ is closest to the group median, with its positive Grad-CAM (CAM), IG and occlusion (Occ.) maps, cropped to the breast. Cyan: lesion (solid) and 5-mm zone (dashed); $+$ : map peak. Below each map: $\lambda_{E}$ (bits; floor about $-$ 6.7) and hit or miss. Bar: 5 cm.

Saliency Map Evaluation

Saliency map evaluation measures whether visual attributions faithfully, stably, and usefully reflect model decisions and human-relevant evidence.

See all 36 papers

Faithful Faithfulness Evaluations: Challenges & Pitfalls Learned from a Breast MRI Case Study
Sep 22, 2026Peachapong Poolpol, Henrik H. J. Detjen, Eike PetersenPerturbation-Based Feature Attribution Saliency Map Evaluation
Fraunhofer Institute for Digital Medicine MEVIS, Bremen, Germany · Deggendorf Institute of Technology, Deggendorf, Germany
Can saliency maps flag chest radiograph classifier errors at a new hospital? An external validation study
Nov 3, 2026Aashish Chapagai, Angad Singh Gurtatta, Erik CambriaSaliency Map Evaluation Explainable Medical Image Analysis
College of Computing and Data Science, Nanyang Technological University, Singapore, Singapore.
From Density to Biopsy Decisions and Malignancy Prediction: A Benchmark Study of Multimodal Large Language Models Against Radiologists in Digital and Contrast-Enhanced Mammography
Sep 13, 2026Ali Abbasian Ardakani, Afshin Mohammadi, Taha Yusuf Kuzan +11Medical VLMs Mammography Image Classification
Department of Medicine, Faculty of Medicine and Dentistry, Danube Private University, Krems an der Donau, Austria · Department of Radiology, Faculty of Medicine, Urmia University of Medical Science, Urmia, Iran · Department of Radiology, Yeditepe University, Istanbul, Türkiye +10

Sep 22, 2026J/K move · Enter open · S save

Faithful Faithfulness Evaluations: Challenges & Pitfalls Learned from a Breast MRI Case Study

Peachapong Poolpol, Henrik H. J. Detjen, Eike Petersen

Fraunhofer Institute for Digital Medicine MEVIS, Bremen, Germany · Deggendorf Institute of Technology, Deggendorf, Germany

Perturbation-Based Feature Attribution Saliency Map Evaluation Medical Image Benchmarks

Open paper AskPDF arXiv

Saliency maps are widely used to explain deep learning predictions in medical imaging, yet visually plausible explanations do not necessarily reflect a model's true decision process and may therefore mislead clinicians. We investigate this problem using a Vision Transformer-based breast MRI classifier trained on the ODELIA Breast MRI Challenge dataset and evaluate multiple saliency methods, including Last-layer Attention, Attention Rollout, Grad-SAM, Gradient Attention Rollout, GMAR, Grad-CAM, and HiResCAM. Our study highlights two often-overlooked challenges in perturbation-based faithfulness evaluation. First, method rankings depend strongly on the perturbation strategy, varying across intensity-based perturbations and transformer-based attention masking. Second, benchmarking saliency methods requires distinguishing between class-specific and class-agnostic explanations. To enable fair comparisons, we introduce non-class-specific variants of gradient-based methods and evaluate both settings separately. Across protocols, Grad-CAM and Gradient Attention Rollout consistently emerged as the strongest class-specific methods, although their relative ranking depended on the evaluation design. These findings expose important limitations of current saliency-based explainability approaches and highlight the need for more robust and standardized evaluation frameworks for trustworthy clinical AI systems.

Abstract

Saliency Map Evaluation

Explore similar work

Faithful Faithfulness Evaluations: Challenges & Pitfalls Learned from a Breast MRI Case Study

Can saliency maps flag chest radiograph classifier errors at a new hospital? An external validation study

From Density to Biopsy Decisions and Malignancy Prediction: A Benchmark Study of Multimodal Large Language Models Against Radiologists in Digital and Contrast-Enhanced Mammography