Statistical textural features for detection of microcalcifications in digitized mammograms

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Clustered microcalcifications on X-ray mammograms are crucial indicators for early breast cancer detection.
• Texture analysis offers a method for identifying these microcalcifications in digitized mammograms.

Purpose of the Study:

• To compare the efficacy of a novel surrounding region-dependence method against conventional texture analysis techniques for detecting clustered microcalcifications.
• To evaluate the performance of these methods in classifying regions of interest (ROIs) as containing microcalcifications or normal tissue.

Main Methods:

• Comparative analysis of texture-analysis methods including spatial gray-level dependence, gray-level run-length, gray-level difference, and the proposed surrounding region-dependence method.
• Extraction of textural features from ROIs.
• Classification of ROIs using a three-layer backpropagation neural network.
• Evaluation of classifier performance via receiver operating-characteristics (ROC) analysis.

Main Results:

• The surrounding region-dependence method demonstrated superior classification accuracy compared to conventional methods.
• The proposed method also exhibited better computational complexity.
• ROC analysis confirmed the enhanced performance of the surrounding region-dependence method.

Conclusions:

• The surrounding region-dependence method is a more effective approach for detecting clustered microcalcifications in mammograms.
• This method offers improved accuracy and efficiency for computer-aided diagnosis of breast cancer.