An SVM classifier to separate false signals from microcalcifications in digital mammograms

Area of Science:

• Medical Imaging
• Machine Learning
• Computer-Aided Diagnosis

Background:

• Digital mammography is crucial for early breast cancer detection.
• Automated detection systems aim to improve diagnostic accuracy and efficiency.
• Microcalcifications are key indicators of early-stage breast cancer.

Purpose of the Study:

• To evaluate the effectiveness of Support Vector Machine (SVM) classifiers for detecting clustered microcalcifications in digital mammograms.
• To compare the performance of SVM against Multi-Layer Perceptron (MLP) classifiers in a false-positive reduction context.
• To assess the impact of training dataset size on classifier performance.

Main Methods:

• Implementation of an automatic detection system utilizing SVM for microcalcification identification.
• Feature extraction from detected signals to differentiate between microcalcifications and false positives.
• Comparative analysis of SVM and MLP classifier performance using Area Under the Curve (Az) values.
• Evaluation across varying sizes of training datasets.

Main Results:

• SVM classifiers achieved slightly superior results (Az=0.963) compared to MLP (Az=0.958) with large training datasets.
• SVM demonstrated significantly better performance (Az=0.952) than MLP (Az=0.918) when using reduced training datasets.
• SVM classifier configuration was found to be simpler than that of MLP.

Conclusions:

• SVM classifiers are a feasible and effective tool for automated detection of clustered microcalcifications in digital mammograms.
• SVM offers advantages in terms of accuracy and ease of implementation, particularly in scenarios with limited training data.
• The findings support the integration of SVM into computer-aided diagnosis systems for mammography.