What is Image Quality?

A strong consensus is emerging in the medical-imaging community on the appropriate definition of image quality. In this view, image quality is defined in terms of the performance of some observer on some clinically relevant task.

The task of interest can be divided generically into classification and estimation tasks. The simplest classification task is detection of a nonrandom signal on a nonrandom background (usually called SKE/BKE, for signal known exactly, background known exactly). More complicated classification tasks involve random signal and backgrounds or multiple image classes. Estimation tasks are concerned with extraction of numerical information from the images. For both kinds of tasks, the goal is to design the imaging system or reconstruction algorithm in such a way as to optimize the performance of the chosen observer on the task of interest.

For estimation tasks, the term “observer” refers to a computer algorithm used to analyze the image and report numerical values for one or more parameters of interest. We have concentrated on estimation algorithms that are optimal in some well-defined sense. They include the generalized Wiener estimator, a Bayesian approach that incorporates prior knowledge of the parameter being estimated, and a Gauss-Markov estimator, which is the best linear unbiased estimator.

For classification tasks, the observer can be a human or a mathematical model. The performance of a human observer is measured by psychophysical studies and analyzed by ROC (receiver operating characteristic) curves. The mathematical-model observers that we consider are either ideal observers or human models. Ideal observers, which set an upper limit to the performance of any observer on the task of interest, include the true ideal or Bayesian observer, which sets an absolute upper limit to task performance, and the ideal linear or Hotelling observer, which is optimum among all observers constrained to perform only linear operations on the data. Human-model observers attempt to account for certain known characteristics of the human visual system. The success of this endeavor is judged by how well the model observers predict the outcome of psychophysical experiments.

Our Mission

The mission of most medical imaging research groups is ultimately to make “better” images. Thus, it is incumbent on us to give a clear definition of image quality and to establish a methodology for demonstrating that one has indeed improved the quality of medical images. Objective assessment of image quality has long been a major emphasis at the University of Arizona as well as other institutions, and together we hope to develop a repository of image-quality information — both in terms of published literature and freely accessable code. Although we are in the early stages of development, we hope to soon have fully-functional literature and software links that will allow others to use this knowledge in their research which we feel will benefit the medical imaging community as a whole.

Collaborative Institutions

The Image Quality Group is currently made up of researchers from the University of Arizona, Food and Drug Administration (FDA), University of California at Davis, University of Massachusetts Medical Center, and State University of New York — Stony Brook.

Funding Sources

The Center for Gamma-Ray Imaging is a Resource Grant funded by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) of NIH.

Grant number: 5 P41 RR14304-020
Principal Investigator: Harrison H. Barrett, Ph.D.
Additional Links: National Institute of Biomedical Imaging and Bioengineering (NIBIB) Homepage

External Websites

Charles Metz’s ROC Software

ROC Literature

Medical Image Perception Society (MIPS)