Explainable Artificial Intelligence in Mammography: A Systematic Review of Methods, Evaluation Practices, and Clinical Readiness
Erklärbare künstliche Intelligenz in der Mammographie: systematische Übersichtsarbeit zu Methoden, Evaluationspraxis und klinischer Einsatzbereitschaft
Diagnostics · 06.05.2026 · 16 (9) · 1412
DOI: 10.3390/diagnostics16091412 ↗Auf einen Blick
14 Studien zeigen: XAI in der Mammographie wird häufig visuell plausibel dargestellt, aber nur selten extern validiert oder klinisch geprüft.
Fragestellung
Welche XAI-Methoden werden in der Mammographie eingesetzt, und wie werden ihre Erklärungen auf Validität, Robustheit und klinischen Nutzen bewertet?
Methode
Systematische Übersichtsarbeit nach PRISMA 2020; Suche in MEDLINE/PubMed, Embase, Scopus, Web of Science Core Collection und Cochrane Library vom 1. Januar 2015 bis 15. Januar 2026; strukturierte narrative Synthese.
Ergebnisse
14 Studien. 10 zu Detektion oder Läsionsklassifikation, 4 zu Risiko- oder Outcome-Prädiktion. Keine Reader- oder Workflow-Studien. Pointing-Game-Scores für Grad-CAM, Grad-CAM++ und Eigen-CAM: 0,30 bis 0,41.
Einschränkungen
Heterogene Literatur; fünf Studien nur tier-1 plausible Erklärungen, sieben tier-2 intern evaluiert, zwei tier-3 extern geprüft.
Schlussfolgerung der Autor:innen
Die Mammographie-XAI ist von visuell plausiblen, aber inkonsistent validierten Erklärungen geprägt. Externe Validierung und klinikerzentrierte Evaluation sind selten.
Kernaussagen
- XAI in der Mammographie war meist nur visuell plausibel belegt.
- Keine Studie berichtete Reader- oder Workflow-Studien.
- Pointing-Game-Scores lagen nur bei 0,30 bis 0,41.
Deutsche Zusammenfassung: KI-gestützt aus dem Original-Abstract erstellt; alle Zahlen wurden automatisch mit dem Original abgeglichen. Maßgeblich ist der Original-Abstract.
Kurz gefasst
Übersichtsarbeit in der Zeitschrift Diagnostics, online erschienen am 06.05.2026. Verfasst von Filippo Pesapane und 9 weiteren Autor:innen. Beteiligte Einrichtung: Breast Imaging Division, Radiology Department, IEO European Institute of Oncology IRCCS, 20141 Milan, Italy. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Background
Explainable artificial intelligence (XAI) is increasingly proposed to improve trust in mammography-based artificial-intelligence systems, but the validity and clinical readiness of published explanations remain unclear. We aim to systematically review XAI methods applied to mammography and synthesize how explanations are evaluated for validity, robustness, and clinical usefulness.
Methods
We conducted a systematic review according to PRISMA 2020. MEDLINE/PubMed, Embase, Scopus, Web of Science Core Collection, and the Cochrane Library were searched from 1 January 2015 to 15 January 2026. Two reviewers independently screened records and extracted data; disagreements were resolved by consensus with a third reviewer. Included studies used mammography as the primary input and reported an explicit explanation or interpretability mechanism. Because the literature was methodologically heterogeneous, we performed a structured narrative synthesis and an adapted XAI-specific appraisal of explanation claims, quantitative evaluation, external validation, human-factor assessment, and reporting transparency.
Results
Fourteen studies were included. Ten studies addressed detection or lesion classification and four addressed risk or outcome prediction. Primary XAI families were interpretable-by-design architectures (6/14), post hoc saliency or attribution methods (5/14), and feature-level explanation methods (3/14). Five studies remained at tier-1 qualitative plausibility only, seven reached tier-2 internal quantitative explanation evaluation, two reached tier-3 external or cross-dataset interpretability assessment, and none reported reader or workflow studies. In the dedicated mammography saliency benchmark, Pointing Game scores for Grad-CAM, Grad-CAM++, and Eigen-CAM ranged from 0.30 to 0.41, indicating only modest lesion-pointing reliability despite acceptable classifier performance.
Conclusions
Mammography XAI remains dominated by visually plausible explanations that are inconsistently validated. The literature is moving toward task-aligned and intrinsically interpretable designs, yet external validation and clinician-centered evaluation remain rare. Future studies should pre-specify explanation claims, use task-appropriate quantitative metrics, report explanation robustness under distribution shift, and test whether explanations improve human decision-making.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Breast Imaging Division, Radiology Department, IEO European Institute of Oncology IRCCS, 20141 Milan, Italy.
- Applied Research Division for Cognitive and Psychological Science, European Institute of Oncology, IRCCS, 20141 Milan, Italy.
- Department of Oncology and Hemato-Oncology, University of Milan, 20122 Milan, Italy.
ORCID-Profile: Filippo Pesapane, Anna Rotili, Luca Nicosia, Roberto Grasso, Enrico Cassano
Zitieren
APA
Pesapane, F., Rotili, A., Penco, S., Dominelli, V., Priolo, F., Marinucci, I., Nicosia, L., Grasso, R., Pravettoni, G., & Cassano, E. (2026). Explainable Artificial Intelligence in Mammography: A Systematic Review of Methods, Evaluation Practices, and Clinical Readiness. Diagnostics, 16(9), 1412. https://doi.org/10.3390/diagnostics16091412
Vancouver
Pesapane F, Rotili A, Penco S, Dominelli V, Priolo F, Marinucci I, et al. Explainable Artificial Intelligence in Mammography: A Systematic Review of Methods, Evaluation Practices, and Clinical Readiness. Diagnostics. 2026;16(9):1412. doi:10.3390/diagnostics16091412
BibTeX
@article{Pesapane2026,
author = {Pesapane, Filippo and Rotili, Anna and Penco, Silvia and Dominelli, Valeria and Priolo, Francesca and Marinucci, Irene and Nicosia, Luca and Grasso, Roberto and Pravettoni, Gabriella and Cassano, Enrico},
title = {Explainable Artificial Intelligence in Mammography: A Systematic Review of Methods, Evaluation Practices, and Clinical Readiness},
journal = {Diagnostics},
year = {2026},
volume = {16},
number = {9},
pages = {1412},
doi = {10.3390/diagnostics16091412}
}
EUVIMED