Automated optimization of JPEG 2000 encoder options based on model observer performance for detecting variable...

Area of Science:

• Medical Imaging
• Image Compression
• Computational Imaging

Background:

• Medical imaging generates large datasets, necessitating efficient compression.
• JPEG 2000 is a proposed standard for image compression.
• Current JPEG 2000 recommendations lack task-based optimization for medical images.

Purpose of the Study:

• To optimize JPEG 2000 encoder settings for medical image compression.
• To improve diagnostic task performance using optimized compression settings.
• To validate optimized settings using various model and human observers.

Main Methods:

• Utilized a genetic algorithm to optimize JPEG 2000 encoder options.
• Employed a non-prewhitening matched filter with an eye filter (NPWE) model observer for optimization.
• Evaluated performance using signal known exactly but variable (SKEV) and signal known statistically (SKS) tasks in X-ray coronary angiograms.
• Compared NPWE-optimized settings with default settings using multiple model observers and human observers.

Main Results:

• NPWE-optimized JPEG 2000 settings significantly improved detection performance for SKEV tasks compared to default settings.
• Performance gains were also observed for the more clinically relevant SKS task.
• Human observers showed up to a 75% improvement in detection performance at high compression ratios (30:1).

Conclusions:

• Task-based optimization of JPEG 2000 encoder settings enhances diagnostic performance in medical imaging.
• The NPWE model observer effectively guided optimization for improved visual detection.
• Optimized JPEG 2000 settings offer substantial benefits for medical image compression, particularly at high compression ratios.