“When you can measure what you are speaking about and express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meager and unsatisfactory kind.”
When 19th-century British physicist and mathematician Lord Kelvin made that declaration, he wasn’t addressing digital radiography (DR), Modulation Transfer Function (MTF) or Detective Quantum Efficiency (DQE).
But he could have been.
As DR continues to make inroads in radiology departments worldwide, manufacturers are jockeying for market position, with each attempting to convince potential buyers to choose their system over the other fellow’s. And much like Lord Kelvin, manufacturers know that numbers can be persuasive.
Often, healthcare facilities go in search of DR technology for reasons of efficiency: They want to stem film loss; increase throughput; streamline workflow; and capitalize on their ability to capture, process, store and distribute images within and beyond their medical complex. They consider user convenience, operational design and – ultimately and always – cost.
But do technological differences make a difference in their purchasing decisions?
Manufacturers are betting that buyers would consider technology, if it could be measured and expressed in numbers so that they “know something about it.” That’s why the accent on MTF and DQE – at least for starters.
MTF and DQE apply to flat-panel detectors for purposes of this article, but as one physicist pointed out, those measurements apply to any kind of imaging system, including CCD-based DR, computed radiography (CR) and film-screen.
Direct and indirect: How similar detectors differ
When flat-panel detectors became commercially available about 1999, their debut capped approximately 10 years of classified research and development efforts in the industry. And while all flat panels share common technical elements, all are not created equal.
Each has a panel of pixel electrodes or storage elements, for example. In indirect conversion detectors, these pixels collect the charge; in direct detectors, they collect the light.
Essentially, direct detectors use amorphous selenium, or a-Se, as a photoconductor, largely due to knowledge of its properties gained through decades-long use as a photoconductor in the photocopying industry. The a-Se, layered onto an array of thin-film transistors (TFT) that serve as little switches, directly converts X-ray photons into an electrical charge.
Indirect detectors, on the other hand, employ a phosphor or scintillator material – typically cesium iodide or a gadolinium agent – that captures the X-ray and converts it to light. The light is then detected by an array of thin-film diodes (TFD) that converts the light to electrical charge.
That extra step – going from X-ray to visible light to electron – earns these detectors the “indirect” label.
“Initially, one might suppose the direct detection technique would be preferable, because any time that you convert one type of quantum into another in a detection scheme you are going to have the potential for introducing additional noise into the image,” remarks James T. Dobbins III, Ph.D., associate professor of radiology and biomedical engineering, at Duke University Medical Center, (Durham, N.C.). “In that case, by virtue of going from X-ray to visible light to electron, you have introduced an extra stage, the visible light stage, that might add additional noise to the image. However, for these devices that have been produced for large-area arrays for chest imaging at higher kilo voltages, it has actually turned out that the performance, to date, in terms of noise, has been better for the indirect than for the direct-detection technique.
“And the reason for that is that any X-ray beam, when intercepted by a detector, contains information at a wide range of spatial frequencies,” Dobbins explains. “The very highest spatial frequencies usually contain very little anatomical information, but they do contain noise. The indirect detection device that has the conversion to visible light actually has a little degree of blurriness compared to the direct-detection device, and that blurriness tends to mask out those higher frequencies of noise so that you are left with not as much noise in the high-frequency area. In that case, the slight case of blurriness from those indirect detection devices is actually a benefit.”
Dobbins acknowledges that his department has agreements with Philips Medical Systems North America (Shelton, Conn.) and GE Medical Systems (GEMS of Waukesha, Wis.) to investigate those companies’ indirect digital detectors. He, personally, also has a research agreement with GEMS to investigate advanced applications with flat-panel devices. The National Institutes of Health (Bethesda, Md.) also has provided Dobbins with grant money for advanced flat-panel-applications research in the fields of dual-energy imaging and digital tomosynthesis.
The data is shown as a function of spatial frequency (f), where higher spatial frequencies correspond to increasingly fine structures in the image. The MTF is a measure of spatial resolution. An MTF closer to 1.0 implies the image looks more like the object. The DQE is a measure of dose efficiency. The closer the DQE is to 1.0, the better the image quality for the dose specified. Direct detectors generally have better MTF, but the a-Se direct detectors generally stop fewer X-rays than the indirect CsI:Tl detectors, thus they have lower DQE.
Andrew D. Maidment, Ph.D., assistant professor of radiology and director of radiological imaging physics at Thomas Jefferson University (Philadelphia, Pa.), agrees with the process that Dobbins describes. Yet, he disagrees with the conclusion.
“Because you have that extra step, you’ve added an inefficiency, and you’ve added some blurring because the light has to escape and the only way it can do that is to scatter,” Maidment says. “In doing that, it spreads laterally; the thicker the screen, the more it will spread, which is an important point. Thicker screens will stop more X-rays so they are more dose-efficient, but the tradeoff is, they blur. A second tradeoff is that X-rays interacting in it don’t transfer all their energy into light, so you’ve lost some signal and that makes your signal-to-noise ratio potentially worse.
“With selenium [direct detectors], those electrons are captured by the electrical field and high voltage, and they get brought down to the array very, very rapidly and they don’t have a chance to spread, so the neat thing with these is that they have basically perfect resolution,” he elaborates. “The resolution in these is determined exclusively by the size of the pixels; if you made pixels half their size, you would get twice as good resolution, whereas in the indirect, resolution is a product of two things: the size of the pixels and the blurring of the phosphorus screen.”
Thomas Jefferson does not own any DR systems for clinical use, Maidment admits, but the school had been a clinical site for direct-detector maker Direct Radiography Corp. (DRC), acquired in 1999 by Hologic Inc. (Bedford, Mass.). That detector, one of DRC’s earliest, is now in the lab. Additionally, Maidment has a DRC-brand panel in his own lab. He also lays claim to having built the first clinical digital mammography system in the world while working on his doctoral degree. That system was remanufactured by Fischer Imaging Corp. (Denver), with the first clinical unit being installed at Thomas Jefferson, he says.
MTF and DQE: The physicists’ alphabet
Martin Yaffi, Ph.D. senior scientist at the University of Toronto, Sunnybrook and Women’s Health Sciences Center (Toronto, Ontario, Canada), says that from a physicist’s point of view, two important evaluations of detector performance are MTF and DQE measurements, but he adds that they are just that – measurements. To date, there are no legal levels or gold standard, that detector manufacturers are required to ? or should meet ? to make good- quality images.
MTF measures image spatial frequency. It is similar to the graphic equalizer on your stereo in that it represents the response of your stereo to different frequencies. In the visual realm, MTF tells you how much high-, mid- and low-frequency you can record. MTF numbers always start high, then drop, as imaged objects become more fine.
Both Dobbins and Maidment point out that selenium-based direct detectors have nearly perfect MTF, which is determined exclusively by their pixel size. As a result, direct detector manufacturers – the most common of which is Hologic (which sells systems to Agfa Corp., Analogic Corp. and Eastman Kodak Co.) – tout their product’s MTF number.
But that’s just one part.
Indirect detectors, by comparison, have lower MTF, influenced by phosphor blurring. At the same time, however, that blurriness favorably influences the signal-to-noise ratio, also known as DQE, resulting in higher rates of DQE. For companies that take the indirect approach – GEMS and Trixell S.A.S. (Moirans, France), among them – DQE is the number to promote.
And that’s just one other part.
Does either part carry more weight?
You be the judge.
Yaffi suggests that, when evaluating a detector’s physical performance, DQE tells a more complete story. It takes into consideration both MTF, which is the signal response, and the Noise Power Spectrum (NPS), which is the noise response. MTF ignores the noise factor, he says.
A Dose of Practicality“You treasure what you measure,” quips Bruce Whiting, Ph.D. “If you can measure DQE, that’s what you discuss.” Whiting, research assistant professor at Mallinckrodt Institute of Radiology (MIR), which serves as the radiology department for Washington University Medical School (St. Louis, Mo.), agrees that Modulation Transfer Function (MTF) and Detective Quantum Efficiency (DQE) have some bearing on detector performance and utility. But he – much like colleague David L. Melson, an electrical engineer and MIR’s project manager for PACS – is of the opinion that digital radiology (DR) technology has yet to “mature.” Whiting worked with Eastman Kodak Co. (Rochester, N.Y.) on its computed radiology (CR) technology for 10 years before coming to Mallinckrodt four years ago. “It certainly has been demonstrated that these systems can provide very nice images,” Whiting observes, “but I think there is a real question about the cost involved, what the durability and maintainability is. The acquisition step is one part of the PACS [picture archiving and communications system]. You input an image, you’ve got to be able to manage it and transmit and store it, you’ve got to be able to interpret to diagnosis from it, so there is a whole display issue that, in my opinion, is currently the weak link in everything.” Melson points out that Mallinckrodt has no flat-panel DR system, even though the Institute – a plum customer in any manufacturer’s portfolio – has been approached by several vendors offering to install a demonstration system. CR, even with its limitations, “is our best option at this point,” he acknowledges. “It is not important for MIR to be one of the first to have DR.” As PACS manager for eight years, Melson is less familiar with MTF and DQE measurements and more concerned with DR specifications that will enable any system to integrate with the different PACS assigned to various sections within radiology: A Siemens Medical Systems Inc. (Iselin, N.J.) system handles CT and MR; a Philips Medical Systems North America (Shelton, Conn.) system manages chest work; an ALI Imaging Systems Corp. (Hartland, Wis.) installation commands ultrasound; and a small GE Medical Systems (GEMS of Waukesha, Wis.) PACS, just recently installed, focuses on mammography. From his point of view, DR is in need of what Melson calls “mechanical, rotational flexibility” to accommodate orthopedic work. He also expects DR to integrate seamlessly with the Mallinckrodt radiology information system (RIS) and to conform to standards coming out of the Integrating the Healthcare Enterprise (IHE) effort, a joint initiative of RSNA and HIMSS. Other important considerations? Well, there is this business of increasing throughput. “These technologies have a differing maximum image throughput rate,” he says. “One, I know, at one point in time, had to refresh their system after an image acquisition at a minimum of 43 seconds – that number sticks out in my mind. That would be considered too slow in certain instances for us,” such as emergency room work, he notes. Whiting contends that MTF and DQE numbers may not make much difference to the user of the system – not because he or she doesn’t appreciate the physics involved, but because people have a way of surpassing manufacturers expectations. “In this field, the quantitative terms have never been convincingly or conclusively demonstrated to link with the actual performance,” he explains. “Again, it’s not just the acquisition,” he adds. “If you are going to read off a monitor, the monitor is very much limited in its output characteristics compared with film on a light box, but people perform adequately; they can do their job on CRTs.” Melson emphasizes that he is only one member of a team of people charged with making the decision to bring DR to MIR; at the same time, he does not shy away from questioning the ability of vendors to service their systems this early in the DR evolution. “We have concerns about serviceability long-term, stability. Is the technology really ready to be a reliable system, and are the vendors prepared to deliver reliable service to those units? There are a number of units out in the field, but, at least to my knowledge, we haven’t seen a high-volume institution like we have where this has been a real big winner. “It is important that we see more visible success stories,” he insists. “There’s the bleeding edge of the leading edge.” |
When things aren’t always what they seem
Now any physicist will tell you that noise complicates any discussion of digital detectors. That’s because there’s noise – and then there’s noise “aliasing.”
In everyday life, the closest example of noise might be your listening to your radio in your car. The farther you get from the radio station, the more static, or noise, you hear. That’s because the signal is getting weaker as you drive farther away.
But noise is constant; everything emits or reflects radio waves. When the radio tries to amplify the signal from the station, it also amplifies the noise and, at a certain point, you begin to hear the noise as static. The noise was always there, but the signal was stronger when you were closer to the radio station, so you did not hear the noise.
Medical imaging accepts that noise principle and goes one better.
When a direct detector forms an image, exclusively with pixels, very high frequencies that a radiologist normally would not see disguise themselves as, or take the alias of, lower frequencies that a radiologist will see. Thus, potential gains in resolution due to nearly perfect MTF and the absolute absence of blur, may be offset by noise aliasing.
Indirect detectors, due to the blurriness created in that “extra” visible light stage, may produce an image that is a little less sharp. But that blur also tends to cap the level of noise frequencies that a radiologist will see. For those detectors, then, noise aliasing does not apply – at least not in large-field applications like chest imaging, according to Dobbins. Mammography may be a different story, he allows.
“There are two factions in physics today,” Maidment states. “One thinks aliasing is a problem. The other doesn’t.”
Maidment is one who doesn’t.
If the tradeoff for a higher-resolution image is noise interference that masks itself as lower frequencies, Maidment wholeheartedly believes that extra resolution is preferable.
“I would rather know there is something there than nothing there,” he insists. “Better to give me the information and let me work it up. Let me have the information and let me figure it out. That is my philosophy.
“The fact is, most of medical imaging is noise-limited,” Maidment continues. “Most of medical imaging is limited by the clutter of the body – you have bone overlying your lungs, for example. That is much worse than the noise from the electronics.”
“Typically, the thing that is important for the human observer is the signal-to-noise ratio. It tells me how much information content is there in what I’m looking for,” counters Dobbins. “Noise basically robs me of the ability to see information content. What’s what our brain and our eye pick out – how much signal there is relative to the noise in the background that masks it. The DQE is more than just that ratio, but that ratio is the fundamental part of the DQE.”
Can you keep the noise down?
In physics, noise is often referred to as “an inability to quantify things precisely.” Much the same can be said for the ongoing, “noisy” debate surrounding MTF and especially DQE measurements.
While there are no legal standards for either measurement, as Yaffi indicated, MTF is the better known of the two. There are publications addressing MTF, particularly as applied to film.
But the clamor surrounding DQE is deafening.
The America Association of Physicists in Medicine (AAPM) has assembled Task Group 16 and charged it with calculating NPS, one of the components of DQE. Maidment chairs that committee.
Meanwhile, an international effort is under way regarding DQE. The International Electrotechnical Commission (IEC), based in Switzerland, includes technical committees working to establish DQE standards.
The IEC is one to two years away from setting a standard, he estimates.
A young and restless technology
In the marketplace, meanwhile, manufacturers continue to monitor and market their current technology while also exploring advanced applications for their particular breed of detector.
“It’s a very young technology, but I think these devices are going to provide the foundation on which we do X-ray imaging for the next 100 years,” reflects Dobbins. “I don’t think there is an awful lot you can do to improve on these, except making incremental improvements: a little bit of DQE, maybe smaller pixels, but that’s all polishing. There are two things that I think make them significant. One is that they provide better image quality than film could, but also they open the door for a wide array of advanced applications that were not possible with film or even with CR [computed radiography] for that matter.
“To be honest,” he adds, “I think what will happen in the next five years is you are going to see every manufacturer have a flat-panel device because everyone has got to have one. They are the hot item.”
