Clockwise from top left: Cutaway of Direct Radiography Corp.’s 2560 x 3072 detector element array; Canon Medical Systems’ sensor panel for the CxDI-22 system; Trixell’s Pixium 4600; GE Medical Systems’ DR flat panel
Comparing the different types of digital radiography technology is getting to be a daunting task. As more systems become available to the market, the radiologist’s and administrator’s understanding of the intricacies of these extremely complex products becomes more difficult. In our February issue, the “Implementing Digital” series looked at some of the differences among the CCD-based DR systems. Fresh off the heels of the 2000 meeting of the International Society for Optical Imaging (SPIE) in San Diego, this piece examines the differences, similarities, benefits and pitfalls of flat-panel DR technologies through the eyes of three top researchers in the DR field. Find out what questions to ask when you’re buying a DR system, as well as what technology to budget for in the future.
James T. Dobbins III, Ph.D.
Associate Professor of Radiology and Biomedical Engineering
Duke University Medical Center
James Dobbins has been involved in digital radiographic imaging research for more than 15 years, including both computed radiography and flat-panel detectors. Most notably, Dobbins says he has developed a number of DR algorithms, including dual-energy imaging and digital tomosynthesis. Dobbins also has done work in the area of the scientific analysis of image quality (both theoretical and experimental).
What projects have you participated in which involve the development or analyzing of a DR detector?
We have recently measured the modulation transfer function (MTF), noise to signal ratio and detective quantum efficiency (DQE) of the first GE [Medical Systems] 41cm by 41cm flat-panel detector for chest radiography. We found its performance to be excellent. We also have a flat-panel device from another vendor that we will evaluate later this year.
What are the differences in the flat-panel detectors available today?
There are two basic varieties: the “indirect” and “direct” detection methods.
The indirect method typically uses a photodiode array detecting light emitted from a scintillator, whereas the direct detection method uses a photoconductor that directly absorbs X-rays. The industry seems to be standardizing on structured cesium iodide (CsI) for the indirect method, and amorphous selenium for the direct method, although there is considerable effort in finding alternate materials for the direct devices.
To your knowledge, which array technology provides the highest DQE and how is that achieved?
I believe the best overall single measure for looking at performance of devices is DQE(f), since it reflects exposure efficiency and overall signal-to-noise ratio at various frequencies. To date, the best DQE(f) for radiographic applications at moderate to high exposures has been the indirect method. These devices combine the relatively good resolution and absorption properties of CsI. The direct devices, typically using selenium, have shown somewhat less DQE, perhaps due to noise aliasing from the very high MTF, but they are improving.
Which technology has shown the most effectiveness in terms of spatial resolution?
The direct detection devices clearly have the best potential spatial resolution since their MTF is limited mainly by the pixel size. But that extra-high MTF at high frequencies can be a detriment, too, since it may lead to aliasing. The indirect devices, while somewhat blurrier, have a band-limit imposed by the CsI which may prove useful for reducing noise aliasing. But overall, really, both indirect and direct detection devices have shown exquisite detail in the images, largely due to the significantly improved DQE at high frequencies relative to film or computed radiography.
What applications beyond radiography and mammography can we expect to see in DR?
Tomosynthesis is an exciting advanced application for flat-panel detectors. With higher-quality flat-panel devices now available, there is strong potential for clinical implementation of tomosynthesis. We are evaluating tomosynthesis for its ability to be a cost-effective and low-dose method of improving detection of lung nodules using a technique we have developed called Matrix Inversion Tomosynthesis that eliminates the blurry artifacts from outside the plane of interest.
Both direct and indirect detection devices should be applicable to tomosynthesis, assuming that they can read out the images fast enough to collect all of the projection images necessary within one patient breath hold.
Both indirect and direct devices are being developed for mammography. There is the potential perhaps for better performance at small pixel sizes with the direct detection method, due to electric field shaping giving almost 100 percent fill factor. However, the optimum size of pixels for mammography is still a topic of some debate, and there are indirect detection devices with 100 micron pixel sizes that seem to perform well also.
Both indirect and direct detection devices are being developed for fluoroscopy, although I do not believe any device has yet shown the performance required for optimum fluoroscopy.
Other exciting areas of advanced application work that also hold promise for use with flat-panel devices include dual-energy imaging and temporal subtraction.
What is the best method for increasing image area?
There are various strategies in both camps for whether to tile panels for increased image area or not. There are large-area detectors available in both direct and indirect varieties that have a tiled layout (either 2 or 4 panels). I am only aware of one detector, which is of the indirect variety, that has a single untiled panel. There are potential advantages to either approach: the untiled panel clearly avoids the problem of physical joining, dead space and algorithmic stitching of adjacent panels, while the tiled detector may potentially hold costs down by providing improved manufacturing yield.
Larry Antonuk, Ph.D.
Associate Professor of Radiation Physics
Radiation Oncology Department
University of Michigan Medical School
Ann Arbor, Mich.
Larry Antonuk’s work in flat-panel DR goes back to the technology’s earliest beginnings in the mid to late 1980s. For much of that time he has collaborated with Robert Street of Xerox’s Palo Alto Research Center (PARC) and the two jointly conceived of the idea of using thin film transistor technology to make active matrix devices that are the basis of today’s flat-panel DR systems. Today, Antonuk leads a 12-person research group at Michigan funded primarily through grants from the National Institutes of Health.
Your history in the research of flat-panel detectors gives you a unique perspective on the history of the technology’s development. How do you think it has progressed since its inception?
Wonderfully. In the beginning we didn’t have any imagers, we just had ideas. The example has been given to us by the flat-panel display industry, which around 1987 had very high ambitions to make large area liquid crystal displays (LCDs) built from active matrix technology. Eventually we formed our first array, which was 64 by 40 pixels in size. That has progressed to where there are devices now with many millions of pixels.
We in the detector development areas are getting a huge boost from the fact that so much money has gone into the development of LCDs. If it weren’t for LCDs, this industry would not exist. The resources required for creating the large area electronics were vast but the incentives were very large for display markets.
Can you explain the difference between direct and indirect detection DR technology?
Direct detection of the radiation involves the X-ray coming in and creating electron hole pairs in a photo conductor material and the hole pairs are then swept away to gather them in a storage capacitor. In an indirect detection device, the X-rays come in and they generate light. The light can’t be gathered but the light propagates through the scintillator until it emerges and exits into a photo-sensitive cell in the pixel. That will generate an electron hole pair – one electron hole pair per optical photon. And then those electron hole pairs are gathered up.
It’s indirect in the sense that one doesn’t directly gather the signal that’s generated by the X-rays. Rather, there’s an in-between step, which has some consequences as to how these devices operate.
Are the two technologies drastically different?
No. I think it’s important to recognize that the two technologies have much more in common than they have different. In both technologies there is a matrix of active matrix switches. A network of lines that turn on and off the transistors and lines to carry off the signal. At every intersection point there is a switch, which in most of the devices is a thin film transistor. That tends to be a superior switch.
I think, perhaps for reasons of competitiveness early on, part of the industry started to look at direct detection for commercialization. Perhaps it was advantageous to portray that there was a larger difference between the technologies than there is.
It’s understandable because product differentiation is so important to them. In fact, these devices share a very common parentage and are really siblings. In terms of the different types, it’s a good demarcation ? how they detect the radiation. How they differ is essentially by how they detect the radiation.
Where they vary is that the direct detection devices, a photoconductive layer is laid in electrical contact with the electrode of that capacitor in the pixel and the signal is stored in the capacitor. In the indirect detection devices, that capacitor is linked to or is itself a photosensitive element.
Which technology provides the highest DQE?
To my knowledge, the indirect detection devices have demonstrated the highest DQE over the largest frequency ranges. That is what has been presented at the conferences I am aware of.
The differences are not vast between direct and indirect. At the same time it should be said the direct detection device that has been commercialized has a DQE at least as good as film. The FDA approval for that device incorporated the concept that it’s DQE be at least as good as film.
The potential is that the direct detection devices should be able to improve their DQE somewhat. Can they be as good as indirect detection? Perhaps. That answer is further qualified as to what spatial resolution domain you want to deal with.
At the moment, direct detection devices suffer in DQE because of a very complicated effect called noise power aliasing, a direct result of the high spatial resolution in amorphous selenium itself. Ironically while this appeared to be a huge advantage to the direct detection devices, it turns out it actually decreases the DQE at higher spatial frequencies. We now think it is possible that there are tricks or procedures to overcome the noise power aliasing problem.
Are there differences in the actual images produced by systems using direct and indirect methods?
The direct detection devices present an image that looks a little different. It’s hard to quantify what that means because the images we see from the manufacturers include some image processing algorithms. It’s hard to know what’s the detector and what’s the image processing algorithm.
However, it is safe to say that neither device shows an advantage in the range of 100 to 200 micron pixel pitch because both devices are limited by the size of the pixel. The pixel size is the limiting factor for the configurations that have thus far been demonstrated for radiographic imaging.
What factors should a potential DR customer look for to distinguish the technologies?
Depending on the application, a potential buyer should inquire as to the quality of the images at low exposures. The intrinsic electronic noise at low exposures will come out. It is important for a customer to see the quality of the images at the lowest exposure that they want to use it at.
They should find out how much a device will drift. The drift properties could affect calibration and procedures need to be outlined. They should ask how often do these devices need to be calibrated? How long does the calibration hold up? Do they calibrate automatically? What kind of quality assurance is necessary, and on what time scale to insure that the calibration procedures which produce offset and gains correction factors?
The potential customer should ask about ghosting. The electronic structure of the material is amorphous silicon, which is a material that can hold a lot of trapped charge. As a consequence, amorphous silicon in the transistors and sensors for indirect devices and for direct detection devices, and to some extent amorphous selenium, will exhibit ghosting or carry over of charge.
What future applications do you see flat-panel technology moving into?
In principle, the direct detection products should be able to do fluoroscopy well but that’s not been demonstrated in a sense that’s practical. High DQE results need to be demonstrated at clinically relevant exposures for fluoroscopy. That’s not been done yet. They’re both capable of doing breast imaging too.
Do you see any cost differences between the two technologies?
Cost issues are about equivalent. I don’t think you’ll see any major difference. The direct detection devices have a simpler active matrix but then they need a more expensive coating. The indirect detection devices have a more complicated active matrix but then the scintillator is a little more direct to put on.
John Rowlands, Ph.D.
Senior Scientist and Professor
Sunnybrook and Women’s Health Sciences Centre
University of Toronto’s Departments of Medical Imaging and Medical Biophysics
Toronto, Ontario, Canada
John Rowlands has been working on DR projects for nearly 20 years. He says that 15 years ago he started work on amorphous selenium-based systems and began to concentrate on active matrix or flat-panel amorphous selenium technologies in 1990. His work on flat panels has been in conjunction with Wei Zhao, an assistant professor at the State University of New York at Stony Brook. The two are currently performing joint work on the optimization of direct flat-panel detectors and development of new clinical applications for digital mammography, as well as investigating advanced active matrix structures for enhancement of signal to noise ratio.
While it has been established there are two types of flat-panel detectors (direct and indirect), what do you see as the differences within the direct detection group?
There are two major variations in direct conversion detectors. The first involves using a higher atomic number semiconductor as the X-ray transducer to replace amorphous selenium. This has three theoretical advantages: higher X-ray absorption, higher X-ray to charge conversion gain, and lower operating potential. However, they have countless practical disadvantages – which may eventually be overcome.
The second approach was the first commercialized and uses a dielectric layer for high voltage protection of the TFT panel, but limits the readout to radiographic (non-realtime) mode as there is a need to reset the panel optically. Other methods, which can be used also fluoroscopically, are to incorporate a high voltage protection device at each pixel or the most recent innovation is to reverse the polarity of the selenium layer which leads to an automatic protection mode.
To your knowledge, which detector technology provides the highest DQE and how is that achieved?
This is not yet a topic that has been fully resolved, but was a subject of lively debate at the recent SPIE meeting. However, we can say that the CsI phosphor-based systems with their very high atomic number for the primary transducer and very high density have a larger value of DQE at zero spatial frequency than selenium-based systems. On the contrary, the selenium-based systems often have the highest DQE at very high spatial frequencies.
Which technology has shown the best spatial resolution?
Undoubtedly, selenium-based systems are superior in this respect.
Which technology lends itself to use in digital fluoroscopic applications?
Here the balance between the systems may favor CsI-based systems due to their very high DQE. However, in my opinion, there is a need for a breakthrough technology to fully implement flat-panel imagers for fluoroscopy as the signal size from both existing systems is inadequate.
The issue of lag in flat-panel imagers is greatly misunderstood. Both systems in principle and in practice have sufficiently rapid response that they can be used for fluoroscopy. There appears to be a widespread, but fallacious, belief that there is a lot of lag in the amorphous selenium layer.
In fact, from the lag point of view, the most challenging situation is what is called dual-mode operation – when the system is used to take a digital radiograph and then is immediately used to obtain a fluoroscopic sequence. Then it is possible to see the radiographic image burned into the detector. It seems that, contrary to popular wisdom, the indirect detector has the most problem and we believe that is due to the amorphous silicon photodiodes.
Does any one technology provide a more cost-effective solution in the long-term?
In the long run it should be possible to manufacture selenium-based systems more economically than CsI/photodiode systems. The CsI layer has to have a well-defined microstructure which is difficult to make, and expensive. The active matrix is more specialized and available only from suppliers who have to bear the capital cost of the foundry based on their sales of sensor panels.
Are there any supply issues between these technologies?
Manufacturers making indirect conversion devices need to set up their own fabrication facility for the active matrix array due to the need for amorphous silicon layers not used in conventional liquid crystal display fabrication. However, manufacturers of direct conversion sensor panels have typically outsourced the manufacture of their active matrix panels to amorphous silicon foundries financed to make huge volumes of active matrix arrays for displays. Thus, the latter manufacturers have access to panels without the capital outlay for their own facility and therefore have more potential for cost-effectiveness. But, on the other hand, do not have the control of having manufacturing in-house.
The preparation of the CsI layer is difficult and the commercial suppliers are limited.