What Is Contrast Resolution?

Contrast resolution is the ability of an imaging system to distinguish between regions that differ only slightly in signal intensity. Where spatial resolution describes how close two features can be before they merge, contrast resolution describes how similar two features can be in brightness or density before they become indistinguishable. The two are separate axes of image quality, and a system can be strong on one and weak on the other: a high-resolution radiograph may still fail to separate a low-density lesion from surrounding soft tissue.

The term is used in two related senses. In the display and encoding sense it refers to the number of distinguishable intensity levels a system can represent, which is bounded by bit depth, so an 8-bit channel offers 256 levels and a 12-bit medical image offers 4,096. In the detection sense, which dominates practical assessment, contrast resolution refers to the smallest intensity difference an observer can reliably detect against background noise, a quantity limited by physics and dose rather than by the number of available code values.

Contrast, Noise, and Detectability

Detectable contrast cannot be defined without reference to noise, because the same intensity difference is obvious in a quiet image and invisible in a grainy one. The working figure of merit is the contrast-to-noise ratio, computed as the difference in mean signal between a region of interest and its background, divided by the standard deviation of the background. As this ratio rises, low-contrast features become visible; below roughly the Rose criterion of about 5, detection becomes unreliable. Work on the use of contrast-to-noise ratio to evaluate CT image quality examines how well this simple metric predicts actual low-contrast detail detectability, and where it diverges from what observers can see.

Measurement and Model Observers

Contrast resolution is measured with phantoms containing arrays of low-contrast inserts whose diameter and density difference are known, such as the modules in the ACR CT and Catphan test objects. Historically a human reader scored which inserts were visible, a subjective procedure with poor reproducibility across readers and sessions. Model observers replaced much of that scoring: the channelized Hotelling observer and non-prewhitening variants compute a detection statistic from the image data and correlate well with human performance. A method for low-contrast detectability and dose optimization using model observers shows how the approach is applied to compare iterative reconstruction algorithms, which change noise texture in ways that a simple contrast-to-noise ratio can misrepresent.

Display and Perceptual Limits

The final stage in the chain is the display, where contrast resolution depends on luminance range, ambient light, and the mapping from pixel value to luminance. Because human contrast sensitivity is not linear in luminance, medical displays are calibrated to a perceptually uniform grayscale standard so that equal steps in pixel value produce approximately equal steps in perceived brightness. Quality control procedures set out in AAPM Task Group 270 on display quality assurance include grayscale ramp patterns for evaluating effective bit depth and low-contrast targets for checking luminance response, since an uncalibrated monitor can discard contrast that the acquisition system worked to preserve.

Applications

Contrast resolution has applications in a range of fields, including:

• Computed tomography and radiographic quality control
• MRI and ultrasound protocol optimization
• Radiation dose reduction and reconstruction algorithm evaluation
• Industrial radiography and nondestructive testing
• Remote sensing and satellite radiometric assessment
• Display calibration and diagnostic workstation acceptance testing
• Machine vision inspection under low-contrast conditions