Mass detection in digitized mammograms using two independent computer-assisted diagnosis schemes

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis (CAD)
• Radiology

Background:

• Mammography is crucial for early breast cancer detection.
• Computer-assisted diagnosis (CAD) systems aim to improve mammogram interpretation accuracy.
• Independent CAD schemes may have complementary strengths and weaknesses in mass detection.

Purpose of the Study:

• To evaluate the potential of combining two independent CAD schemes for mass detection in mammograms.
• To investigate the impact of logical 'or' and 'and' operations on sensitivity and specificity.
• To determine if combining CAD results can improve diagnostic performance beyond individual schemes.

Main Methods:

• Two distinct CAD schemes were applied to a database of 428 digitized mammograms containing 220 verified masses.
• Scheme 1 (CAD-1) utilized Gaussian bandpass filtering and multilayer topographic feature analysis.
• Scheme 2 (CAD-2) employed a five-stage search for suspicious regions.
• The performance of each scheme and their combination using logical 'or' and 'and' operations were compared.

Main Results:

• CAD-1 achieved 96% sensitivity with a 0.79 false-positive rate per image.
• CAD-2 achieved 94% sensitivity with a 1.69 false-positive rate per image.
• The 'or' operation combined sensitivity reached 100% but increased false positives to 2.07 per image.
• The 'and' operation reduced false positives to 0.4 per image, but sensitivity dropped to 90%.

Conclusions:

• Combining independent CAD schemes, analogous to double reading, can enhance either sensitivity or specificity.
• The choice of logical operation ('or' vs. 'and') allows tailoring the combined CAD output to clinical priorities.
• This approach offers a flexible strategy to optimize mass detection performance in mammography.