Hybrid Deep Learning EfficientNetV2 and Vision Transformer (EffNetV2-ViT) Model for Breast Cancer Histopathological Image Classification

Abstract

Breast cancer remains a leading cause of death among women, highlighting the urgent need for effective detection methods. In recent years, AI-based techniques, including computer vision, machine learning, and deep learning, have gained significant popularity in the field of medical imaging. The healthcare industry has witnessed remarkable progress due to these AI techniques, particularly in the early detection of cancer, which can greatly impact patient outcomes and survival rates. This research introduces a new approach to identifying breast cancer by combining two advanced computer technologies: EfficientNetV2 and vision transformer, using a specific dataset called BreakHis. EfficientNetV2 is praised for its quick processing and efficient use of resources, making it an excellent tool for initially identify important information in the data. We utilize three variants of EfficientNetV2 (small, medium and large) in order to discern the crucial features. Afterwards, this information is processed by a transformer, a type of model excellent at classifying or sorting data, to determine if breast cancer is present. Our experiments show that this method, especially when using the largest version of EfficientNetV2 paired with the vision transformer, is highly effective in accurately identifying breast cancer. It reached an impressive accuracy of nearly 99.83% when deciding between two possible categories and 98.10% when distinguishing among eight categories. These promising results depict that combining these two technologies could be a powerful way to improve breast cancer detection accuracy.

Keywords

• Breast cancer
• deep learning
• EfficientNetV2
• transformer models
• medical imaging
• classification