Enhanced MobileNet with multi-scale feature fusion for automated breast cancer histopathology classification

Abstract

Abstract Accurate and efficient diagnosis of breast cancer from histopathological images remains a major challenge in clinical practice due to subjective interpretation, inter-observer variability, and labor-intensive manual examination. To address these limitations, this work introduces a transfer learning–based framework for automated breast cancer classification using the Breast Cancer Histology Images (BACH) dataset. Several pre-trained deep architectures—including MobileNet, ResNet variants, EfficientNet, and Vision Transformers—were evaluated and extended with a Multi-Scale Feature Fusion (MSFF) module to capture morphological heterogeneity across spatial resolutions. Among these, the Enhanced MobileNet (E‑MobileNet) with MSFF outperforming recent state‑of‑the‑art models and achieving a classification accuracy of 95%, precision of 95%, recall of 94%, and F1‑score of 96%. The framework was further validated on the BreaKHis dataset across multiple magnifications, achieving an average accuracy of 90.6%. These results confirm the robustness and generalization capability of the proposed model for practical clinical deployment in digital pathology.