Applying computer-assisted detection schemes to digitized mammograms after JPEG data compression: an assessment

Area of Science:

• Medical Imaging
• Digital Signal Processing
• Radiology

Background:

• Digital mammography is crucial for breast cancer screening.
• Computer-assisted detection (CAD) systems enhance mammogram interpretation.
• Image compression techniques like JPEG are used to reduce file sizes for storage and transmission.

Purpose of the Study:

• To evaluate the impact of Joint Photographic Experts Group (JPEG) image compression on the performance of CAD systems.
• To assess the effects on the detection of masses and microcalcification clusters in digitized mammograms.

Main Methods:

• 952 digitized mammograms were compressed using a JPEG-compatible scheme at five different levels.
• A previously developed CAD scheme, optimized for noncompressed images, was applied to the reconstructed compressed images.
• Performance was compared against the original noncompressed digitized mammograms.

Main Results:

• For mass detection, no significant differences in true-positive or false-positive rates were observed between compressed and noncompressed images.
• The CAD scheme detected 80% of masses with less than one false-positive per image across all compression levels for masses.
• Significant performance degradation was noted for microcalcification cluster detection (P < .001) with increasing compression ratios (17:1 to 62:1), showing reduced sensitivity (4%-10%) and increased false-positive rates (91%-140%).

Conclusions:

• JPEG compression did not adversely affect the performance of the CAD scheme for detecting masses on mammograms.
• JPEG compression significantly impaired the CAD scheme's ability to detect microcalcification clusters.