• AI
  • Molecular Imaging
  • CT
  • X-Ray
  • Ultrasound
  • MRI
  • Facility Management
  • Mammography

Mathematical model approximates human visual perception of compressed images

Article

Because it is so closely correlated with human observer perception, a model based on just noticeable differences (JND) is better able to predict image quality than traditional quantitative measures, according to a Monday morning informatics presentation.

Determination of optimal lossy compression levels for CT and CR has been difficult. The JNDmetrix visual discrimination model correlates extremely well with radiologists' assessments of image quality, in comparison with those traditionally used by physicists, such as peak signal-to-noise ratio. (Provided by K. Siddiqui)

Because it is so closely correlated with human observer perception, a model based on just noticeable differences (JND) is better able to predict image quality than traditional quantitative measures, according to a Monday morning informatics presentation.Currently, there is a disconnect between what industry uses and what humans use to determine image quality, said Dr. Bruce Reiner, director of radiology research at the VA Maryland Health Care System.Whereas traditionally accepted image quality metrics such as mean square error and peak signal-to-noise ratios (PSNR) are quantitative measures, humans rely on more subjective measures such as overall image quality and lesion dectectibility, Reiner said.The JNDmetrix model values are better predictors of image quality because they can simulate the physiology of the human visual system, said Reiner, who presented study findings on behalf of lead researcher Dr. Khan M. Siddiqui.To test the JND visual discrimination model, 11 radiologists examined 80 CT and computed radiography images on 3-megapixel LCD monitors. The images had undergone a range of JPEG compression from lossless up to 60:1.The radiologists judged overall quality and readability of the images on a scale of one to 10. They then normalized those reader scores and correlated them with PSNR and JND values calculated for the images.As expected, the researchers found that measures of image acceptability were inversely proportional to the level of image compression, Reiner said.Additionally, normalized reader scores were extremely highly correlated to JND values for both CT (-0.9) and CR (-0.91). Peak signal-to-noise ratios did not correlate as well to human observations, with values of 0.78 for CT and 0.63 for CR.The differences between PSNR and JND correlation to normalized reader scores was statistically significant.Traditionally, researchers have relied on receiver operating characteristic studies to quantitatively determine image quality, Reiner said. The JND model could simplify determination of acceptable image compression levels and help decide whether lowered radiation doses will yield acceptable images. "It could eventually replace the time-intensive and complicated ROC studies," he said.For more information from the diagnostic imaging online archives:3D CT outperforms 2D in compressing thin slices Method takes on determination of optimal image compression ratio Compression research expands from wavelets to Glicbawls

Recent Videos
Current and Emerging Insights on AI in Breast Imaging: An Interview with Mark Traill, Part 1
Addressing Cybersecurity Issues in Radiology
Computed Tomography Study Shows Emergence of Silicosis in Engineered Stone Countertop Workers
Can an Emerging AI Software for DBT Help Reduce Disparities in Breast Cancer Screening?
Skeletal Muscle Loss and Dementia: What Emerging MRI Research Reveals
Magnetoencephalopathy Study Suggests Link Between Concussions and Slower Aperiodic Activity in Adolescent Football Players
Radiology Study Finds Increasing Rates of Non-Physician Practitioner Image Interpretation in Office Settings
Assessing a Landmark Change in CMS Reimbursement for Diagnostic Radiopharmaceuticals
Addressing the Early Impact of National Breast Density Notification for Mammography Reports
2 KOLs are featured in this series.
Related Content
© 2024 MJH Life Sciences

All rights reserved.