Fusing global context with multiscale context for enhanced breast cancer classification

Area of Science:

• Oncology
• Computer Vision
• Medical Imaging

Background:

• Breast cancer is a leading cause of cancer death in women.
• Accurate classification of breast cancer from histopathological images is crucial for effective treatment.
• Current Convolutional Neural Networks (CNNs) have limitations in capturing global and multi-scale features, impacting classification accuracy.

Purpose of the Study:

• To develop an advanced fusion model for accurate breast cancer classification.
• To overcome the limitations of CNNs by integrating global and multi-scale feature extraction.
• To enhance the diagnostic performance for breast cancer detection using histopathological images.

Main Methods:

• A fusion model integrating Vision Transformer (ViT) for global features and Atrous Spatial Pyramid Pooling (ASPP) for multi-scale features was developed.
• An attention mechanism was incorporated into the model architecture.
• A five-stage image preprocessing technique was applied to histopathological images.

Main Results:

• The proposed fusion model achieved 100% accuracy on the BreakHis dataset at 100X and 400X magnification.
• Classification accuracies of 99.25% at 40X and 98.26% at 200X magnification were recorded.
• The model demonstrated robust performance across different magnification levels.

Conclusions:

• The fusion model effectively classifies breast cancer from histopathological images.
• The integration of ViT and ASPP enhances feature extraction capabilities for improved accuracy.
• This model presents a dependable tool for proficient breast cancer classification and early detection.