Computerized image analysis: texture-field orientation method for pectoral muscle identification on MLO-view...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate pectoral muscle identification is crucial for mammogram analysis.
• Existing methods may lack automation or robustness across imaging modalities.

Purpose of the Study:

• To develop and evaluate a novel texture-field orientation (TFO) method for automated pectoral muscle identification on mammograms.
• To assess the TFO method's accuracy and adaptability to different mammogram types.

Main Methods:

• Developed a gradient-based directional kernel (GDK) filter and texture analysis for orientation mapping.
• Implemented ridge point extraction, validation, and shortest-path finding for boundary tracking.
• Evaluated the TFO method on digitized film mammograms (DFMs) and full-field digital mammograms (FFDMs) using overlap and distance metrics.

Main Results:

• The TFO method achieved a mean percent overlap area (POA) of 95.0 ± 3.6%.
• High accuracy was observed, with 91.5% of detections exceeding 90% POA.
• Distance metrics showed high concordance with radiologist delineations (e.g., 99.4% within 5 mm average distance).

Conclusions:

• The automated TFO method accurately detects pectoral muscles on DFMs.
• The TFO algorithm demonstrates robustness and adaptability to FFDMs without retraining.