Breast cancer histopathology image classification using transformer with discrete wavelet transform.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Accurate breast cancer diagnosis from histopathology images is critical for treatment.
• Deep learning models excel at image classification but often overlook frequency domain features, limiting performance.
• Existing methods primarily focus on spatial features, neglecting valuable frequency information.

Purpose of the Study:

• To propose a novel deep learning network, DWNAT-Net, for enhanced breast cancer histopathology image classification.
• To integrate Discrete Wavelet Transform (DWT) with Neighborhood Attention Transformer (NAT) to leverage both frequency and spatial information.
• To improve the accuracy and efficiency of automated breast cancer diagnosis.

Main Methods:

• Developed DWNAT-Net, incorporating Discrete Wavelet Transform (DWT) for multi-frequency band decomposition and Neighborhood Attention (NA) within a Transformer architecture.
• DWT extracts frequency features while preserving spatial details by iterative filtering and downsampling.
• NA focuses attention computation on local neighborhoods for efficient dependency modeling.

Main Results:

• DWNAT-Net achieved high image-level recognition accuracy on benchmark datasets.
• Achieved 99.66% accuracy on the BreakHis dataset.
• Achieved 91.25% accuracy on the BACH dataset, outperforming existing state-of-the-art methods.

Conclusions:

• The proposed DWNAT-Net effectively integrates frequency and spatial features for superior breast cancer histopathology image classification.
• The model demonstrates significant improvements in diagnostic accuracy, offering a promising tool for early cancer detection.
• DWNAT-Net presents a competitive and advanced approach compared to current state-of-the-art methods in computational pathology.