Deep Learning-Based Automated Breast Cancer Detection from Histopathology Images: Benchmarking CNN and Transformer Architectures
Year: 2026
Type: conference-paper
Abstract: Breast cancer continues to be one of the leading health challenges worldwide, where accurate and timely diagnosis is crucial for effective patient management. Microscopic examination of tissue specimens plays a central role in confirming the presence of malignant abnormalities; however, manual interpretation of biopsy slides is labor-intensive and may vary between specialists. Recent advances in machine learning have enabled the development of automated systems capable of assisting the analysis of medical images. This study investigates the effectiveness of different deep neural network architectures for identifying breast tumor categories from histopathology images. Experiments were conducted on the BreaKHis dataset using the 200× magnification subset for an eight-class classification task. Six representative models were evaluated, DenseNet121, EfficientNetB3, ConvNeXt, Vision Transformer (ViT), Swin Transformer, and MaxViT, under the same experimental configuration to ensure consistent comparison. The experimental results show that the Swin Transformer achieved the highest accuracy of 0.8713 and recall of 0.8711, while ConvNeXt obtained the best precision (0.8889) and AUROC (0.9714). These findings indicate that attention-driven architectures can effectively utilize broader spatial information in histopathology images, while advanced convolution-based networks remain strong alternatives for medical image recognition tasks.
Institution: BRAC University
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204469954