Optimizing parameters for computer-aided diagnosis of microcalcifications at mammography

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Mammography is crucial for breast cancer screening.
• Distinguishing benign from malignant microcalcifications remains a challenge.
• Computer-aided diagnosis (CAD) systems can aid in image analysis.

Purpose of the Study:

• To optimize the selection of mammographic features for improved discrimination of benign from malignant clustered microcalcifications.
• To develop and evaluate a classification scheme based on optimized features.

Main Methods:

• A computer-aided diagnosis (CAD) system extracted 13 quantitative features from digitized mammograms.
• Stepwise discriminant analysis identified the most effective features for classification.
• A classification scheme was built using these features and evaluated with receiver operating characteristic (ROC) analysis.

Main Results:

• Six of 13 CAD-extracted features were selected, yielding an ROC area (Az) of 0.98 with 98% sensitivity, 83.64% specificity, and 91.79% accuracy.
• The classification scheme showed significantly higher performance for women aged 50 or older (Az=0.99) compared to younger women (Az=0.96).
• Adding patient age as a variable did not significantly improve the scheme's performance.

Conclusions:

• Stepwise discriminant analysis effectively optimized a classification scheme for microcalcifications using six key features.
• The developed scheme demonstrates high diagnostic performance, especially for older age groups.
• While patient age did not significantly enhance classification, the study highlights the potential of CAD systems in mammographic analysis.