Sliding window based deep ensemble system for breast cancer classification

Area of Science:

• Oncology
• Computer Science
• Medical Imaging

Background:

• Breast cancer poses a significant global health challenge, with early detection crucial for reducing mortality rates.
• Accurate classification of breast cancer from histopathological images remains a complex task.
• Deep learning methods show promise for high-performance breast cancer screening.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for classifying breast cancer into eight distinct types using histopathological images.
• To enhance the accuracy of breast cancer detection through an ensemble method combining multiple pre-trained convolutional neural networks (CNNs).

Main Methods:

• A sliding window technique was applied to histopathological images from the BreakHis database.
• Four pre-trained CNNs (GoogleNet, AlexNet, ResNet50, DenseNet-201) were utilized in an ensemble approach.
• The method was tested on images with magnifications of 40x, 100x, 200x, and 400x.

Main Results:

• The proposed ensemble technique achieved a high accuracy of 99.3325% in classifying breast cancer histopathological images.
• The system demonstrated effective classification across all tested magnification levels.
• The results are comparable to those of recent studies in the field.

Conclusions:

• The developed ensemble deep learning method offers a highly accurate and effective solution for breast cancer classification.
• This approach has the potential to significantly improve the accuracy and efficiency of breast cancer screening and diagnosis.
• Further research can explore the application of this method to larger datasets and other cancer types.