CMAF-Net: Cross-Modal Attention Fusion with Information-Theoretic Regularization for Imbalanced Breast Cancer Histopathology

This repository contains the official implementation of CMAF-Net, proposed in

CMAF-Net: Cross-Modal Attention Fusion with Information-Theoretic Regularization for Imbalanced Breast Cancer Histopathology

CMAF-Net combines convolutional and transformer streams with cross-modal attention and information-theoretic regularization to improve classification of breast cancer histopathology images under severe class imbalance.

The code includes:

• cmaf_model.py: CMAF-Net architecture.
• dataset.py: dataset loaders and transformations.
• train.py: training and evaluation script.
• utils.py: metrics and helper functions.
• sweep.py: optional hyperparameter sweep script.
• splits/: train/validation/test splits for IDC and BreakHis.
• notebooks/: analysis and visualisation notebooks.
• figures/: figures generated from experiments (e.g., confusion matrices).

Note: At this stage (as the manuscript is under review) the repository is kept private; source code, trained weights, and experiment scripts will be released publicly upon acceptance of the manuscript.