Multi-Class Breast Cancer Classification Using Ensemble of Pretrained models and Transfer Learning.

Area of Science:

• Medical imaging analysis
• Machine learning in healthcare

Background:

• Breast cancer is a leading cause of cancer-related deaths globally, particularly among women.
• Early detection significantly improves patient outcomes and survival rates.
• Computer-aided diagnosis (CAD) systems have evolved with advancements in machine learning and deep learning.

Purpose of the Study:

• To develop an effective breast cancer detection method using ensemble learning.
• To perform both 2-class and 8-class classification for breast cancer diagnosis.
• To address challenges of imbalanced datasets in medical classification tasks.

Main Methods:

• An ensemble of pre-trained models was utilized to handle data imbalance.
• The proposed method was evaluated on 2-class and 8-class classification tasks.
• Research utilized Google Cloud Platform with 2 Nvidia Tesla V100 GPUs for implementation.

Main Results:

• Achieved 98.5% training accuracy and 89% test accuracy for 8-class classification.
• Attained 99.1% training accuracy and 98% test accuracy for 2-class classification.
• Demonstrated high performance in distinguishing between different breast cancer classes.

Conclusions:

• Ensemble learning effectively improves breast cancer classification accuracy.
• Dataset imbalance can lead to misclassifications, particularly in specific classes.
• Future work may involve increasing dataset size or exploring alternative methodologies to further enhance model performance.