Causal and Counterfactual Explainable AI in Breast Cancer Imaging: A Clinically-Constrained Review Toward Trustworthy Diagnosis

DOI

10.2991/978-94-6239-799-6_5How to use a DOI?

Keywords

Breast Cancer Imaging; Explainable Artificial Intelligence (XAI); Causal Inference; Counterfactual Explanations; BI-RADS; Deep Learning; Medical Image Analysis

Abstract

Breast cancer has been one of the major causes of cancer induced death among women across the globe, with early and proper diagnosis by the use of imaging tests such as mammography, ultrasound and magnetic resonance imaging (MRI), being fundamental in enhancing survival. Over the last few years, deep learning-based computer-aided diagnosis systems are reported to have high predictive accuracy with a reported accuracy of over 90 percent and area under the curve (AUC) values of up to 0.94 on benchmark data sets such as the CBIS-DDSM and BUSI dataset. The black-box character of such models however inhibits their application in clinical settings because of lack of interpretability, transparency and trust. The review gives an exhaustive and clinically-based assessment of explainable artificial intelligence (XAI) techniques in breast cancer imaging with specific attention to the systems of causal inference and counterfactual explanations. In comparison to traditional saliency-based methods (e.g., Grad-CAM, SHAP), which provide correlational information, causal and counterfactual methods allow reasoning about the effects of the application of theoretical or counterfactual change scenarios, and the factors that have influenced model predictions. We review the existing approaches systematically with limitations on them to produce clinically infeasible or anatomically inconsistent explanations. The review provides a solution to this gap by highlighting the new paradigm of clinically-constrained explainability, specifically, with the BI-RADS (Breast Imaging Reporting and Data System) as a domain-specific prior to direct counterfactual generation. Having a limited search space, which is formed by the medically valid combinations of features, allows the BI-RADS constrained methods to substantially enhance the plausibility and interpretability of the explanations, and align AI outputs with radiological practice. Moreover, we present and integrate new progress in causal stability analysis which measures the stability of explanations across imaging applications, scanners and acquisition settings, and micro-causal causal pathways that offer patient-specific interpretability and permit systematic study of model failures. Data sets, performance measures, and validation procedures also owe comparison in this review, and as it can be seen the existing models have high accuracy (around 90%) but few studies evaluate clinical plausibility and consistency across domains. Among the issues leading to these problems, we single out the bias of the dataset used, the scarcity of standardized evaluation framework to explain the model, and the insufficient integration of clinical knowledge in the design of the model.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Yashika Yadav
AU  - Ajay Kumar Sharma
AU  - Mayank Patel
PY  - 2026
DA  - 2026/09/30
TI  - Causal and Counterfactual Explainable AI in Breast Cancer Imaging: A Clinically-Constrained Review Toward Trustworthy Diagnosis
BT  - Proceedings of the International Conference on Emerging Trends and Technologies in Applications of Computer Science (ICETTACS 2026)
PB  - Atlantis Press
SP  - 54
EP  - 62
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-799-6_5
DO  - 10.2991/978-94-6239-799-6_5
ID  - Yadav2026
ER  -