Dual-Level Explainable CNN–ViT for Breast Cancer Histopathology Classification

Abstract

Breast cancer diagnosis from histopathological images remains a critical yet challenging task due to high intra-class variability and reliance on expert interpretation. Although deep learning models have achieved promising results, many existing approaches suffer from limited interpretability and insufficient modelling of global contextual dependencies. This study proposes a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) with Vision Transformers (ViTs) to capture local texture features and long-range contextual relationships jointly. To enhance model transparency, a dual-level explainability mechanism is incorporated by combining gradient-based and modelagnostic techniques. This multi-level explainability provides complementary insights into model behaviour and improves the reliability of predictions for clinical applications. The combination of explainable AI and deep learning could enhance transparency and trust in AI-driven breast cancer classification systems, thereby aiding clinical decision-making.The proposed model is evaluated on the BreakHis-400x dataset, achieving 94.11% accuracy, 98.18% area under the curve (AUC), and 95.76% F1- score. Overall, the proposed framework effectively integrates hybrid feature learning with multi-level interpretability, offering a reliable and transparent approach for breast cancer classification from histopathological images. The model shows strong potential for deployment in computeraided diagnostic systems.