.png){kind=icon}
.png){kind=icon}
Return
Enhancing breast cancer detection using EffiAtt-ViT: an EfficientNet-attention-vision transformer fusion framework
.png){kind=icon}
PRE
AI
A
M
Y
J
DOI：10.1016/j.bspc.2026.111239.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Breast cancer (BC) is increasing rapidly. Early detection of BC can improve treatment and decrease the mortality rate. However, BC detection at early stages is challenging due to the complex histopathological tissue structure, variations in staining, and magnification levels. Also, the high morphological similarity among different breast lesion subtypes makes accurate diagnosis difficult. Previous studies have primarily relied on convolutional neural networks (CNNs), which effectively capture local tissue features but often fail to model long-range contextual dependencies. To address these limitations, we propose a novel EfficientNet-Attention-Vision Transformer (EffiAtt-ViT) framework for BC detection. We use EfficientNet to extract discriminative multi-scale local tissue features. After that, the Convolutional Block Attention Module (CBAM) is used to enhance salient channel and spatial feature representations. The Vision Transformer (ViT) is used to capture long-range contextual dependencies among refined feature embeddings. After that, we perform feature fusion by integrating the locally discriminative features extracted by EfficientNet with the globally contextual representations learned by the ViT. This enables the model to maintain detailed cellular morphology while capturing the overall tissue structure. To validate the model performance, we use two publicly available datasets, known as Breast Cancer Histopathological Image Classification (BreakHis) and the Breast Cancer Histology images (BACH) dataset. We obtained an accuracy of 98.45% using BreakHis and 95.39% using the BACH dataset, outperforming closely related works. This work contributes to the existing literature by presenting an optimized EffiAtt-ViT framework that integrates multi-scale feature extraction, attention-guided learning, and global contextual modeling for breast histopathology image classification.
Journal
IF:
4.9
Papers:
9.7K
Citations:
2.4W
Organization
Cited Papers