BreastHistoNet: A Efficient Breast Cancer Histopathological Image Classification Using Multiscale Features and...

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Accurate breast cancer (BrCan) diagnosis relies on histopathological image analysis.
• Current deep learning models, while effective, are computationally intensive and memory-demanding, limiting their clinical utility.
• There is a need for efficient and accurate automated BrCan classification tools.

Purpose of the Study:

• To develop a lightweight deep learning model for automatic BrCan histopathological image classification.
• To achieve performance comparable to state-of-the-art methods while minimizing computational complexity and memory footprint.
• To provide a practical tool for resource-constrained clinical settings.

Main Methods:

• Integration of Depthwise-Dilated-Multiscale-Pointwise (DDMP) blocks for multiscale feature extraction.
• Utilization of Discrete Wavelet Transform (DWT) and Squeeze-and-Excitation (SE) blocks for feature recalibration.
• A dual-stream architecture combining DWT subbands and max-pooling outputs, followed by SE blocks.
• Development of the BreastHistoNet model featuring DDMP-SE blocks, Global Average Pooling, and dense layers with GELU activation.

Main Results:

• BreastHistoNet demonstrates superior efficiency with a model size of 7.47 MB, 0.63 M parameters, and 6.50 G FLOPS.
• Achieved high classification performance on the BreaKHis dataset: 95.48% accuracy, 95.61% precision, 95.46% specificity, 95.46% recall, and 95.48% F1-score.
• Ablation studies confirmed the impact of epochs, activation functions, and batch sizes on model performance.

Conclusions:

• BreastHistoNet offers a highly accurate and computationally efficient solution for BrCan classification.
• The model's low complexity and minimal memory usage make it suitable for deployment in clinical environments.
• This lightweight approach facilitates the integration of AI in aiding BrCan diagnosis.