Scaling multi-instance support vector machine to breast cancer detection on the BreaKHis dataset

Area of Science:

• Computational pathology
• Machine learning in oncology
• Biomedical image analysis

Background:

• Breast cancer is the most common cancer in US women, necessitating early diagnosis for better outcomes.
• Automated histopathological classification using machine learning aids early breast cancer detection.
• Existing machine learning models often lack scalability for large datasets.

Purpose of the Study:

• To develop a novel, scalable machine learning model for breast cancer histopathological classification.
• To address the limitations of current models in handling large-scale datasets.
• To identify abnormal tissue segments in histopathological images.

Main Methods:

• Introduction of the Primal-Dual Multi-Instance Support Vector Machine (MIM-SVM).
• Development of an efficient optimization algorithm that bypasses traditional quadratic programming and least-squares methods.
• Application of the MIM-SVM to the public BreaKHis dataset.

Main Results:

• The proposed method demonstrates computational efficiency and scalability for large datasets.
• Promising prediction performance was achieved in histopathological classification.
• The model effectively identifies tissue segments indicating abnormalities.

Conclusions:

• The Primal-Dual MIM-SVM is a computationally efficient and scalable solution for breast cancer histopathological classification.
• This approach holds potential for improving early breast cancer diagnosis through automated analysis of large image datasets.
• The developed software is publicly available for further research and application.