Computerized quantitative evaluation of mammographic accreditation phantom images

Area of Science:

• Medical Imaging
• Radiology
• Image Analysis

Background:

• Mammography accreditation requires consistent image quality assessment.
• The American College of Radiology (ACR) phantom is used for quality control.
• Manual scoring of phantom images can be subjective and time-consuming.

Purpose of the Study:

• To develop and evaluate an automated scoring system for ACR mammographic accreditation phantom images.
• To compare the performance of the automated system against human expert scoring.

Main Methods:

• The automated method involved background subtraction, region of interest determination, Mahalanobis distance classification for fibers and masses, template matching for specks, and rule-based scoring.
• Fifty-one phantom images from different facilities were analyzed.
• Human scoring was performed by a medical physicist and two radiologic technologists.

Main Results:

• The automated system achieved 90% accuracy for fibers, 80% for masses, and 98% for specks in meeting ACR criteria.
• A significant association was found between computer and observer scores for microcalcifications (p<5%).
• No significant association was found for masses and fibers between computer and observer scores.

Conclusions:

• The developed automated method offers a stable assessment of test object visibility in mammographic phantom images for ACR accreditation.
• The system demonstrates high accuracy, particularly for speck detection.
• Further improvements are needed for the accurate scoring of fiber and mass objects.