Published at ICONIP 2024 — International Conference on Neural Information Processing
CaDT-Net combines Cascaded Deformable Convolutions with Transformer attention for breast cancer classification from histopathological images, achieving 97.79% validation accuracy on the BreakHis dataset.
| Model | Accuracy |
|---|---|
| ResNet-50 | 94.2% |
| EfficientNet-B3 | 96.8% |
| ViT-B/16 | 97.1% |
| CaDT-Net (Ours) | 97.79% |
BreakHis — 7,909 histopathological images, 2 classes (Benign / Malignant)
pip install -r requirements.txt
python src/train.py --data_dir data/breast_cancer --epochs 30@inproceedings{gobbilla2024cadtnet,
title={CaDT-Net: Cascaded Deformable Transformer for Breast Cancer Classification},
author={Gobbilla, Jaya Krishna},
booktitle={ICONIP},
year={2024}
}
