سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / اولین همایش بین المللی تحول خدمات سلامت: تلفیق مراقبت، درمان، بازتوانی و امیدبخشی

Explainable AI (XAI) and Hybrid Deep Learning Frameworks for Breast Cancer Image Segmentation and Classification: A Systematic Review

نویسندگان :

Fatemeh Ghanaati¹ Alireza Zarei²

1- موسسه آموزش عالی آپادانا 2- دانشگاه بین المللی امام رضا علیه السلام

کلمات کلیدی :

Breast Neoplasms،Deep Learning،Artificial Intelligence،Diagnosis, Computer-Assisted،Image Interpretation, Computer-Assisted

چکیده :

Abstract: Background: Breast cancer remains a leading cause of global cancer-related mortality among women, demanding highly accurate, robust, and interpretable diagnostic tools. While deep learning demonstrates remarkable performance in medical imaging, its “black-box” nature hinders clinical adoption. Therefore, this systematic review evaluates the efficacy of integrating Explainable Artificial Intelligence (XAI) with hybrid deep learning frameworks for breast cancer image segmentation and classification, aiming to bridge the critical gap between computational performance and clinical trust. Method: Following PRISMA 2020 guidelines, a comprehensive search was conducted across PubMed, Scopus, IEEE Xplore, and Web of Science. Boolean search strategies utilized keywords including “Breast Cancer,” “Deep Learning,” “Explainable AI,” “Segmentation,” and “Classification.” Inclusion criteria targeted English, peer-reviewed original articles (2025-2026) explicitly evaluating XAI-integrated hybrid models for breast cancer imaging. Reviews, non-English papers, and studies lacking XAI were excluded. Ultimately, 23 studies met these rigorous criteria for final synthesis. Finding: Data synthesis revealed that hybrid architectures, such as Convolutional Neural Networks combined with Vision Transformers or traditional classifiers, consistently outperform standalone models. XAI techniques primarily Grad-CAM, SHAP, and LIME successfully provided visual and mathematical interpretability for model predictions. However, current research is constrained by a lack of quantitative XAI evaluation metrics, reliance on retrospective unimodal datasets, and high computational complexity. Conclusion: The synergy of hybrid deep learning architectures and XAI establishes a robust, interpretable foundation for breast cancer computer-aided diagnosis (CAD) systems. Overcoming current limitations through multi-modal data integration and standardized XAI evaluation protocols is essential for future clinical translation in precision oncology.