Using a visual discrimination model for the detection of compression artifacts in virtual pathology images

Area of Science:

• Digital Pathology
• Image Compression
• Medical Imaging

Background:

• Telepathology relies on digitized virtual slides, which are large, causing storage and transmission delays.
• Current compression methods (lossless and lossy) have limitations in data reduction or image quality preservation.
• Visually lossless compression offers higher compression ratios but requires adaptive rate control.

Purpose of the Study:

• To evaluate the effectiveness of a visual discrimination model (VDM) for achieving visually lossless compression of virtual slides.
• To determine optimal JPEG 2000 bit rates for visually lossless compression of breast biopsy virtual slides.
• To compare VDM performance against other distortion metrics like PSNR and SSIM.

Main Methods:

• Investigated the utility of a visual discrimination model (VDM) and other distortion metrics.
• Determined threshold bit rates for visually lossless compression using human observers on virtual slide tissue regions.
• Compressed test images to visually lossless thresholds and analyzed VDM-computed just-noticeable difference (JND) metrics.

Main Results:

• VDM-computed JND metrics remained consistent at high percentiles for visually lossless compressed images.
• VDM metrics showed significantly less variability compared to PSNR and SSIM metrics.
• Achieved 5-12 times greater data reduction compared to reversible compression methods.

Conclusions:

• VDM metrics can effectively guide visually lossless compression of virtual slides.
• This approach significantly reduces data size without compromising diagnostic quality.
• Enables more efficient telepathology workflows and data management.