Microcalcification classification assisted by content-based image retrieval for breast cancer diagnosis

Area of Science:

• Medical Imaging
• Machine Learning
• Computer-Aided Diagnosis

Background:

• Mammography is crucial for breast cancer diagnosis, but accurate classification of microcalcifications remains challenging.
• Content-based image retrieval (CBIR) has shown promise in medical image analysis.
• Machine learning models can learn similarity measures from expert observations.

Purpose of the Study:

• To develop a microcalcification classification scheme for breast cancer diagnosis.
• To improve numerical classifier performance by incorporating content-based mammogram retrieval.
• To provide an enhanced "second opinion" for radiologists through improved classification accuracy.

Main Methods:

• A machine learning approach for mammogram retrieval was developed, modeling similarity based on expert observers.
• Retrieved similar mammogram cases were used as references to enhance a numerical classifier.
• An adaptive support vector machine (SVM) was employed to incorporate local proximity information.

Main Results:

• The proposed retrieval-driven approach demonstrated improved classification performance.
• The area under the ROC curve increased from 0.78 to 0.82 with the adaptive SVM.
• Experimental results on a mammogram database validated the effectiveness of the method.

Conclusions:

• Content-based mammogram retrieval can significantly enhance microcalcification classification accuracy.
• The integration of retrieval-driven adaptive learning offers a valuable tool for breast cancer diagnosis.
• This approach has the potential to improve diagnostic support for radiologists.