A deep fusion-based vision transformer for breast cancer classification

Area of Science:

• Digital pathology
• Computational oncology
• Artificial intelligence in medicine

Background:

• Breast cancer remains a leading cause of mortality in women globally.
• Accurate detection of cancerous tissue in histopathology is crucial for diagnosis.
• Existing Convolutional Neural Network (CNN) models struggle to capture complex cellular and staining patterns.

Purpose of the Study:

• To develop an advanced deep learning model for enhanced breast cancer detection in histopathological images.
• To overcome limitations of current methods in feature extraction and image processing.
• To improve diagnostic accuracy through a novel fusion-based approach.

Main Methods:

• Proposed a deep fusion-based vision Transformer model (DFViT) integrating CNNs and transformers.
• DFViT fuses RGB and stain-normalized images to capture both local and global patterns.
• The model was trained and validated on diverse datasets including BreakHis, BACH, and UC.

Main Results:

• DFViT demonstrated superior performance in accuracy, F1 score, precision, and recall.
• The model effectively captures intricate patterns of cell layers and staining properties.
• Achieved state-of-the-art results in histopathological image analysis for breast cancer diagnosis.

Conclusions:

• The proposed DFViT model offers a significant advancement in breast cancer diagnosis using histopathological images.
• Deep fusion of CNNs and transformers enables more effective feature extraction.
• This approach sets a new benchmark for automated analysis in digital pathology.