SegEIR-Net: A Robust Histopathology Image Analysis Framework for Accurate Breast Cancer Classification.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Breast Cancer (BC) poses a significant global health challenge for women.
• Accurate and reliable classification of BC is crucial for early diagnosis.
• Existing classification methods often lack the required accuracy and robustness.

Purpose of the Study:

• To design a robust model for accurate classification of BC histopathology images.
• To leverage advanced segmentation techniques for improved diagnostic accuracy.
• To enhance early detection of breast cancer through precise image analysis.

Main Methods:

• A combined segmentation and classification approach using the Chan-Vese algorithm for region delineation.
• Development of SegEIR-Net, integrating EfficientNet, InceptionNet, and ResNet for classification.
• Application of Bilateral Filtering for noise reduction and utilization of Dense and Dropout layers.

Main Results:

• SegEIR-Net demonstrated superior performance compared to State-of-the-Art (SOTA) methods across multiple datasets.
• Achieved high accuracy rates on the breakHis dataset (up to 98.66%) across different magnifications.
• Attained significant accuracy on the BACH (93.33%) and UCSB (96.55%) datasets, confirming model robustness.

Conclusions:

• The proposed SegEIR-Net framework exhibits robust performance in classifying BC from histopathology images.
• The model's accuracy supports its potential for reliable early diagnosis of breast cancer.
• This approach advances computational pathology for improved cancer detection.