A novel computer-aided diagnosis system for breast MRI based on feature selection and ensemble learning

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Breast cancer is a prevalent disease, necessitating accurate early detection methods.
• Medical imaging, particularly MRI, plays a crucial role in breast cancer diagnosis.
• Automated computer-aided diagnosis (CADx) systems can enhance the efficiency and accuracy of image interpretation.

Purpose of the Study:

• To develop and evaluate an automated computer-aided diagnosis (CADx) framework for breast cancer detection using MRI.
• To improve the accuracy and reduce false positives in breast MRI-based cancer diagnosis.
• To integrate advanced machine learning techniques for robust image analysis.

Main Methods:

• An ensemble of machine learning techniques was employed, including ensemble under-sampling (EUS), Relief feature selection, subspace method, and Adaboost.
• Extracted features included morphological, texture, and Gabor features, combined into meaningful subspaces.
• The framework was tested on a dataset of 438 malignant and 1898 normal/benign breast MRI images.

Main Results:

• The proposed CADx framework achieved an Area Under the ROC Curve (AUC) of 0.9617.
• The system demonstrated superior performance compared to existing state-of-the-art breast MRI CADx systems.
• A significant reduction in the false-positive classification rate was observed.

Conclusions:

• The developed automated CADx framework shows high efficacy in diagnosing breast cancer from MRI.
• The integration of ensemble methods and diverse feature extraction enhances diagnostic performance.
• This approach offers a promising tool for improving breast cancer detection accuracy and efficiency.