Discovering mammography-based machine learning classifiers for breast cancer diagnosis

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Mammography is a key tool for breast cancer screening.
• Accurate interpretation of mammograms is crucial for early diagnosis.
• Machine learning offers potential for enhancing diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a novel machine learning classifier (MLC) for breast cancer diagnosis.
• To optimize MLC configurations using extensive feature vector analysis from mammographic images.
• To improve the accuracy of BI-RADS diagnosis through automated classification.

Main Methods:

• Massive evaluation of machine learning classifier (MLC) configurations.
• Feature vector extraction from segmented regions (lesions or normal tissue) in craniocaudal (CC) and mediolateral oblique (MLO) mammography views.
• Application of image processing and normalization techniques to reduce artifacts and enhance details.
• Utilizing computer clusters for efficient selection of high-performing MLC configurations.
• Validation using 286 cases with radiologist-segmented regions and biopsy-confirmed diagnoses.

Main Results:

• Evaluated approximately 20,000 MLC configurations.
• Achieved a high area under the ROC curve (AUC) of 0.996.
• Demonstrated superior performance when combining feature vectors from both CC and MLO views.

Conclusions:

• The proposed method effectively designs mammography-based machine learning classifiers for breast cancer diagnosis.
• The approach is robust across different data acquisition circumstances.
• Combining features from multiple mammographic views significantly enhances diagnostic performance.