Spatial resolution and MTF can no longer stand alone as gauges of a given system’s diagnostic utility. Instead, a new standard has emerged:

Detective Quantum Efficiency (DQE), the measure of the combined effect of the noise and contrast performance of an imaging system, expressed as a function of object detail.

To understand why DQE is considered by many experts to be the most accurate way to evaluate digital x-ray image quality, let’s consider its components.

Quantum and electronic noise are random variations of signal that can obscure useful information in a diagnostic image; quantum noise arises from variation in the number of x-ray photons detected. Inherent in any digital imaging chain, noise has long been recognized as a parameter that can have a dramatic impact on image quality, which degrades very quickly as noise increases. System noise is expressed by the signal-to-noise ratio (SNR):

Signal         Useful image information
--------   =   ------------------------
Noise         Erratic information

Fig. 1: Signal-to-noise ratio. In digital x-ray systems, as noise decreases, or SNR increases, object detectability increases very rapidly.

Noise is a major limiting factor in object detectability - one that remains constant in a given system unless dose is increased. Low noise is therefore a prerequisite to good image quality at reasonable doses, particularly when viewing small, low-contrast objects.

Another key component of DQE is the radiological contrast performance of an imaging system - its ability to capture an object’s actual contrast.Digital exposure x-ray systems in general offer a wide dynamic range, so they can capture a wide range of signal intensities from the very low to the very high. They also offer very high contrast resolution - that is, they are capable of capturing thousands of shades of gray, far more than the human eye can appreciate. This permits the imaging of areas that would otherwise be under- or overexposed on conventional film.

Fig. 2: Contrast resolution. Because of a digital system’s high contrast resolution, one can enhance the detectability of small objects through window/leveling. Consider the case of an image with a background intensity of 100 and object intensity of 105 – in other words, an object contrast of 5%. With a window of 255 and a level set at zero, the object is nearly indistinguishable from the background (left). As we narrow the window and increase the level, the object becomes more visible (center). By setting the level to that of the background intensity and the window to the relative object intensity – in this case, 5, for 5% – maximum contrast is obtained.

As a result, digital systems with very low noise, very wide dynamic range and very high contrast resolution can improve the detectability of low-contrast objects significantly over film/screen systems. And this low-contrast detectability can be further improved through sophisticated image processing, including automatic contrast enhancement and window/leveling - for instance, by setting the contrast level near that of the back-ground’s, and by narrowing the window to just above and below the object’s signal level (Fig. 2).

Image contrast-to-noise ratio best characterizes “object” detectability.

Fig. 3: SNR and contrast.
Although digital image contrast can be manipulated, this capability cannot overcome a high noise level. Nor can a low noise level compensate entirely for poor contrast capability. As a result, when viewed in a vacuum, neither parameter alone is enough to quantify digital image quality.

To accurately measure a digital imaging system’s performance, we must evaluate the combined effects of noise and contrast performance; these parameters can no longer be viewed in isolation. A system rendering high levels of contrast will not produce diagnostically useful images if its images are very noisy - in others words, if its SNR is low. On the other hand, the diagnostic utility of a system with very low noise will be equally limited if it’s unable to render adequate contrast (Fig. 3).

Both low noise and high-contrast performance are required for superior image quality and object detectability.

The solution: Detective Quantum Efficiency (DQE). Expressed as a function of object detail, or spatial frequency, DQE combines noise and contrast performance into a single parameter that is widely accepted as the measure most representative of digital image quality and object detectability:

Maximizing DQE at all spatial frequencies should, therefore, be the most important goal for designers of digital-detector-based imaging systems - and it was just that for designers of GE’s digital-detector system.

An imaging system’s ability to render the contrast of an object as a function of object detail is traditionally expressed as its Modulation Transfer Function (MTF). But MTF is normally measured under ideal laboratory conditions, using high-contrast objects and high dose while minimizing scatter radiation and noise. Therefore, it is not a reliable indicator of performance in real clinical situations.

Fig. 4: Detector DQEs.
DQEs resulting from various approaches to x-ray imaging (2). Note that most clinically relevant information resides in the low-to-mid frequency range; the very high DQE of GE’s digital detectors results in low-noise images with high object detectability.

What’s more, although MTF can be a useful performance metric for film-based systems offering no post-processing, it is not critical for digital systems. This is because digital post-processing allows one to achieve almost any desired MTF, given adequate SNR.The results are clear when various detectors are compared (Fig. 4).

GE digital detectors typically exhibit a significantly higher DQE than state-of-the-art film/screen, computed radiography, and flat-panel Selenium-based imaging systems, particularly at the low-to-mid spatial frequencies where most clinically relevant information resides. The result is superior object detectability.

Limiting spatial resolution (LSR) is a measure of the spatial frequency at which one can no longer see a high-contrast, structured periodic test pattern under the most favorable test conditions - for instance, in the absence of scatter and focal-spot penumbra and with the use of high dose.

Fig. 5: DQE’s effect on image quality.
Studies comparing film/screen and digital images make the importance of DQE clear. Although the limiting spatial resolution (LSR) of film/screen images (left) is much higher than that of digital images (right), a digital detector’s dramatically higher DQE improves one’s ability to detect small and low contrast objects

These conditions are far removed from typical clinical situations. Yet LSR has traditionally been used as a critical image-quality parameter for film/screen-based systems - and occasionally as an argument for retaining film-based imaging. This is a misconception that must be dispelled if we are to accurately assess digital-detector performance.

A digital system with a relatively low LSR can make up for this limitation by minimizing noise and maximizing contrast performance – in short, by offering better DQE. As a result, even with relatively low LSR, digital images obtained with a high-DQE detector typically offer dramatic improvements in one’s ability to detect small objects (Fig. 5). Therefore, LSR by itself is not an accurate measure of the size of the smallest object a digital imaging system can demonstrate – or of image quality.

This has been confirmed by studies comparing film- and digital-detector-based imaging: Even though film/screen combinations may exhibit LSRs as high as 20 lp/mm, the smallest object that can be detected in an image is usually much larger. Again, this is because, at high spatial frequencies, film exhibits low contrast and high noise - in other words, very low DQE:

Studies indicates that 100 micron spatial resolution may be ideal for full-field digital mammography, according to Loren T. Niklason, Ph.D., formerly a physicist for Massachusetts General Hospital and now a consultant based in Hillsborough, North Carolina.

Fig. 6: Impact of MTF on object detection in noisy images

"It appears that 100-micron resolution is certainly be adequate for general mammography, especially for detection tasks," Dr. Niklason says. "The debate centers on whether we need 50 microns for characterization once a lesion has been detected. So the question becomes, if we see a 200 to 400 micron microcalcification and want to look at its edges to characterize it, will 100-micron resolution enable us to do so?

"So far, it looks like 100 microns is sufficient for this task in the majority of cases."

In the few cases where 50 microns might be useful, he added, a magnification view could be acquired. While this would mean an additional exposure, it would avoid the inevitable trade-offs associated with 50-micron resolution – especially increased noise, the result of capturing fewer x-ray signals per pixel.

"Many researchers agree that when you start adding high levels of noise, there’s very little information left in an image (Fig. 6). To compensate, you’d have to operate at a much higher dose in order to take advantage of that higher resolution."

Improved image quality and object detectability at clinically relevant spatial frequencies are clearly two of the most important potential benefits of a digital-detector system with high DQE. But there are other advantages, as well.

Consider two others of great importance:

Patient dose
is a parameter with direct impact on DQE: DQE is proportional to Image quality/patient dose (Image Quality is approximately related to output SNR, and patient dose is related to input SNR.) This relationship means a digital detector with high DQE should have the potential for improving image quality at the same dose – or for lowering dose without compromising image quality. For example, phantom studies performed with the GE detector have shown that small-object contrast detectability was improved by up to 40% in the 0.2-0.3-mm range, compared to film.

Advanced applications
may be the most important advantage of digital technology - including applications such as dual-energy imaging for selective soft/hard-tissue discrimination, tomosynthesis or 3D imaging for enhanced spatial visualization, and low-dose fluoro for both radiographic and mammographic applications.

Combined with advanced image-processing algorithms, high DQE is the parameter that will make them possible. In fact, accommodating advanced applications has been among the GE detector team’s primary design goals from the start. The uncompromising design of our digital detectors, and their ability to deliver the industry’s leading DQE, make them the best choice for the radiology departments of the future.

Designed from the beginning to help ensure the highest possible DQE, GE digital detectors are already demonstrating their ability to produce images of the highest quality, at the lowest dose, all around the world.

"The detection of small aperiodic objects as small as 100 µm in diameter (like microcalcifications) is determined by limitations in x-quantum statistics rather than by MTF. In particular, it is good DQE at low spatial frequencies which affects signal-to-noise ratio and quantum statistics, and consequently, the size of objects which can be detected."(1)

"It seems clear, that digital mammography systems do not need to reproduce the limiting resolution of the conventional film-screen systems. Practical digital mammography systems can be designed with pixel matrixes significantly smaller than 8,128 x 10,160 [the resolution of a 25-micron film] for an 8" x 10" format and still resolve objects smaller than the conventional film/screen system thanks to better signal-to-noise ratio."(3)

Contrast resolution:
The number of shades of gray that a detector can capture. Flat-panel digital detectors typically offer resolution of 12-14 bits.

Detective Quantum Efficiency (DQE):
An expression of the efficiency of an imaging system’s transfer from its input to its output of both signal and noise, expressed as a percentage. It is the measure most representative of image quality in terms of an observer’s ability to detect objects of interest in an image.

Exposure dynamic range:
The range of exposures over which a detector will generate a usable signal. Because they offer a much wider dynamic range than film-based systems, flat-panel digital detectors are capable of producing images over a very wide range of exposures. This reduces the number of retakes required because of poor technique.

Limiting Spatial Resolution (LSR):
The spatial frequency at which an observer can no longer see a high-contrast, structured periodic test pattern under favorable test conditions - e.g., at high dose, without scatter and focal spot penumbra. Because it does not take into account noise or contrast performance, LSR is not a reliable measure of image quality. Nor is it, by itself, a measure of the smallest aperiodic object a given imaging system can render. Instead, object detectability is controlled primarily by the object’s contrast, and the imaging system’s noise and contrast performance.

Signal-to-Noise Ratio (SNR):
The amount of useful image information (signal) compared to non-useful information (noise).