[Application of a computer-aided detection (CAD) system to digitalized mammograms for identifying...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Healthcare

Background:

• Radiologists miss up to 25% of breast cancers in mammograms, rising to 50% for minimal signs.
• Independent double reading reduces false negatives by 5-15%, but technological advancements offer new solutions.
• Computer-aided detection (CAD) systems are emerging as tools to assist radiologists.

Purpose of the Study:

• To develop and evaluate a CAD system for microcalcification detection.
• To compare the CAD system's performance against human observers.
• To assess the CAD system's utility as a second reader in mammography.

Main Methods:

• Developed a CAD system using ad hoc algorithms and an artificial neural network.
• Utilized a database of 802 digital mammograms from the CALMA project.
• Evaluated images using two experienced radiologists and the developed CAD system.

Main Results:

• The CAD system achieved high sensitivity, identifying 99.3% of microcalcification clusters at thresholds of 0.13-0.14.
• Sensitivity remained high (>82%) for thresholds up to 0.17.
• Lower thresholds increased false positives, with 9-7 false positives/image at optimal thresholds (0.15-0.16).

Conclusions:

• Optimal CAD system performance for microcalcification detection was achieved with thresholds of 0.15-0.16.
• The system offers high sensitivity, crucial for screening, despite lower specificity.
• CAD systems are poised for widespread adoption, providing valuable assistance to radiologists in digital mammography centers.