With the potential for patient dose

Transcription

1 Medical Physics and Informatics Clinical Perspective Walz-Flannigan et al. rtifacts in Digital Radiography Medical Physics and Informatics Clinical Perspective lisa Walz-Flannigan 1 Dayne Magnuson Daniel Erickson Beth Schueler Walz-Flannigan, Magnuson D, Erickson D, Schueler B Keywords: artifact, digital radiography, flat panel, ghosting, lag DOI: /JR Received May 17, 2011; accepted without revision May 24, ll authors: Department of Radiology, Mayo Clinic, 200 First St SW, Rochester, MN ddress correspondence to. Walz-Flannigan (walzflannigan.alisa@mayo.edu). CME This article is available for CME credit. JR 2012; 198: X/12/ merican Roentgen Ray Society rtifacts in Digital Radiography OBJECTIVE. The purpose of this article is to discuss flat-panel digital radiography (DR) artifacts to help physicists, radiologists, and radiologic technologists visually familiarize themselves with an expanded range of artifact appearance. CONCLUSION. Flat-panel DR is a growing area of general radiography. s a radiology community, we are still becoming familiar with these systems and learning about clinically relevant artifacts and how to avoid them. These artifacts highlight important limitations or potential complications in using flat-panel DR systems. With the potential for patient dose reduction, increased workflow, and improved image quality, many radiology practices are purchasing flat-panel digital radiography (DR) systems instead of the more ubiquitous computed radiography (CR) systems. long with potential differences in efficiency, there are differences in acquisition and processing between DR and CR that can create image artifacts that are unique to DR. In using the term artifact, we also include issues, such as image lag, that are related to inherent physical limitations of the detectors. The recommended techniques for minimizing these artifacts can sometimes negate potential efficiency gains of DR compared with CR. Each different DR system model and usage has the prospect for a different manifestation or degree of artifact. Even if vendors use the same physical detector hardware, differences in calibration or image processing may yield different results or risk for artifacts. Establishing standard tests and control limits to identify or prevent clinically relevant artifacts can be challenging. In this article, we present our experiences with different artifacts that have arisen in our work with different DR systems over the course of several years. Our hope is to help expedite the discovery and resolution of DR problems and to provide guidance into what can be controlled or resolved, what is clinically significant, and what might have to be expected and accepted from different DR systems. Many different types of artifacts common to both CR and DR, including under- and overexposure (dose creep), collimation issues, and Moiré artifacts related to grid use have been covered elsewhere [1 6] and will not be readdressed here. Other work has also discussed DR-specific artifacts [6 11]. Materials and Methods The artifact examples in this article are taken from our clinical practice. The systems used include indirect flat-panel imagers with an amorphous silicon (a-si:h) thin-film transistor (TFT) array coupled to either a cesium iodide phosphor or gadolinium oxysulfide phosphor. These systems included a Digital Diagnost (Philips Healthcare), xiom ristos MX (Siemens Healthcare), Definium 8000 (GE Healthcare), and DRX-1 (Carestream Health). Patient image artifacts are presented with the image processing used clinically. Measurements of system performance use raw images or minimal image processing output available from the DR system. Our description of artifacts is organized around specific cases. For each case, we provide clinical image examples, the physical mechanism from which the artifact arises, the potential clinical impact, and suggestions for how it may be addressed or resolved. Results: Specific rtifacts Image Compositing Image compositing refers to combining images from separate exposures. There is often a need to acquire an image longer than the typical length of a CR plate or DR detector (43 cm), particularly for radiographic imaging of the spine or legs. Image compositing, 156 JR:198, January 2012

2 rtifacts in Digital Radiography Fig. 1 Schematic shows movement of x-ray beam and flatpanel detectors in digital radiography system that is set up for image compositing. es are acquired by pivoting the x-ray tube at the location of the focal spot, angled to match the location of the detector for each exposure. In this configuration, any regions of anatomy that overlap are imaged with the same projection geometry so no artifact is created. also called image stitching or image pasting, is accomplished by different methods for CR and DR. To create a long image using CR, multiple cassettes are arranged into a single holder. The plates are overlapped to provide information across plate boundaries. The plates are stationary and a single x-ray exposure is used to expose either two or three plates. fter the imaging plates are read out, the images are combined on the basis of fiducial markings that appear in the images and indicate the location of the overlap. In DR, one of the available methods creates a single image of long anatomy using a synchronous step movement of the x-ray tube and detector. Individual projection images acquired at each tube-detector position are then automatically assembled into a single composite image at the modality workstation on the basis of anatomy or marking from a lead ruler imaged with the patient. s shown in Figure 1, to include all patient anatomy, regions near the top and bottom edge of the detector are included in two adjacent image acquisitions. natomy in the overlap region of the x-ray beam will be projected downward in the upper x-ray tube position and projected upward in the lower x-ray tube position. When the adjacent exposures are combined, the resultant image will overlay and blend the two projections of the same anatomy, potentially resulting in image artifacts. In the scoliosis examination shown in Figure 2, two radiographs were acquired with synchronous step movement of the x-ray tube and detector as previously described. pedicle screw was positioned in the overlapping region of the two projections. When the individual images were stitched together, the different projections of the screw were both visible. Gray-scale blending in the overlap region resulted in different image densities, representing the screw projection. lthough this situation created an image that was clearly artifactual, it is possible that a more subtle superposition of the screw images could occur when the hardware is located at a different depth, thus creating an artifact that mimics the appearance of a loosened screw. Note that if the individually acquired source images before image compositing are reviewed, it is apparent that screw displacement has not occurred. When acquiring composite images, it is recommended that the source images be archived and available for review by the clinician in addition to the full composite image. n alternative acquisition method for image compositing that avoids this artifact is also possible. For this method, source imag- Fig. 2 Image compositing artifact. and B, Frontal full-spine radiographs show upper and lower x-ray tubedetector positions. C and D, Composited image shows overlap region (C) and distorted pedicle screw (D). E, Magnified view of pedicle screw shows blended image of differing acquisition projections. Detector Image Lag and Ghosting The terms lag and ghosting are often used interchangeably. Manifestations of lag and ghosting can be difficult to differentiate because they either appear indistinguishable or both play a role in a given image artifact. In this article, we associate lag with the release of a trapped charge, observable as increased image signal (signal offset), and we associate ghosting with a change in sensitivity (or gain) because of prolonged or excessive exposure over part of a detector (perhaps also from a trapped charge). n exposure must be made to observe ghosting. Flat-panel DR systems generally allow a shorter time interval between exposures compared with cassette-based CR systems. This potential for improved efficiency is one of the major advantages of DR; however, most systems can acquire images at a rate faster than their detectors can actually accommodate. The rapid acquisition of images can result in latent signal from one exposure lingering into the readout of subsequent exposures, producing what appears to be an incomplete erasure of the previous image, known as image lag. The JR:198, January

3 Walz-Flannigan et al. Fig. 3 Image lag artifact. Standing full-length lowerextremity radiograph acquired as image composite from three source images, which are indicated by dotted lines. rtifact in middle source image (lower circle) is caused by area of unattenuated radiation in top source image (upper circle). Detector exposure level estimated to be 0.5 mgy. Solid black ovals obscure patient identifying information. residual lag signal is linked to charge trapping in the a-si:h TFT array of an indirect flat-panel imaging system and is greater in areas that have received high exposure. The image lag in a-si:h TFT arrays has been addressed previously [7, 9, 10]. In a DR system, the signal stored in the a-si:h TFT detector array is read out at regular intervals by the application of readout voltages. Not all of the trapped charge from the first exposure may be released by application of a single readout event, and the residual charge may be released during subsequent readouts, resulting in the lag signal. The amount of residual charge (hence, lag signal) decays logarithmically with the number of readout events [7]. During subsequent exposures, smaller signals (in lower exposure regions) require larger readout voltages that are more efficient at releasing the trapped charge [7]. s a result, lower-exposure regions of an image can suffer greater effects of image lag. This is seen for the two clinical examples shown in this article. ll DR detectors will generate lag signals, but generally most lag will not be clinically relevant, either because it has been minimized by the elapsed time between exposures or another method that compensates for detector dark current. However, image artifacts may occur in clinical practice. The first example of image lag is from a composite image acquired as three individual exposures (Fig. 3). The exposures are taken in quick succession to avoid misalignment because of patient motion. The three source images were acquired within 18 seconds. In the top source image, there is an area outside of the skin line that was exposed to the unattenuated x-ray beam or raw radiation (detector exposure level estimated to be 1.7 mgy). The exposure and readout for the second image occur approximately 6 seconds after the first exposure. ghost of the first exposure appears clearly in this second image (detector exposure level estimated to be 0.5 mgy). The second example of image lag is shown in Figure 4. residual image of the lead markers, indicating laterality and the performing technologist s initials from the previously acquired image, is visible. In the previous image, the lead marker area was exposed to unattenuated x-rays. Because of the presence of the current marker on the other side of the image, potential confusion regarding the actual laterality and technologist may occur. This latent image of the lead marker is easily noticeable because of the sharp lines and high contrast. However, areas of residual signal from other objects could be positioned in a way that could affect image interpretation without appearing to be artifactual. Lag can have a clinical impact if it obscures important features in a subsequent image, mimics a clinical finding, or suggests the wrong laterality. Fortunately, lag signal decreases in time, such that in subsequent exposures the residual image of lead markers will not appear. In addition, lag signals are not typically seen clinically unless an image has a high-contrast object within a region of high exposure and is quickly followed by another image that puts the high-contrast lag signal in an area of lower radiation exposure. To prevent image lag artifacts, we examine the factors that contribute to their appearance. First, because the amount of signal lag increases with exposure, we reduce the amount of unattenuated radiation incident on the detector by reducing the exposure level, collimating or providing bolus filtration. Second, because the amount of signal lag decreases with time and detector readout events, we increase the time interval between exposures or, if possible, decrease the amount of time between readout events (although this is generally not an option available to the end user). Third, if an examination includes multiple images, we acquire the higher-exposure images with larger areas of unattenuated radiation last, allowing time for signal decay before the next patient examination begins. Different DR systems may clear image lag more quickly than others, and some may have features that prevent image acquisition until residual lag signal is depleted. If the primary desired feature of a DR system is increased throughput or efficiency, it is worthwhile to investigate what a realistic acquisition rate may be for a given DR system or what features may be in place to prevent the acquisition of images corrupted by image lag. Flawed Gain Calibration One of the potential advantages of DR (compared with CR) is the correction of imperfections or nonuniformities in the detector or x-ray beam. This correction process generally consists of two steps: a flat-field gain calibration and an offset or dark-noise correction [6]. In this section, we will address artifacts related to gain correction. rtifacts related to a flawed offset correction will be Fig. 4 Image lag artifact. Lateral chest radiograph shows residual image of lead markers from previous acquisition visible over anatomy (square). Solid black oval obscures patient identifying information. 158 JR:198, January 2012

4 rtifacts in Digital Radiography Fig. 5 Flawed gain calibration., Flat-field image shows adjacent plus and minus density areas with blurred margins (arrow). B, Photograph of aluminum beam filtration shows defect (oval). discussed in the next section, Detector Lag and Flawed Offset Correction. Gain calibration is used to compensate for sensitivity variations across the detector. The details of how gain calibration is accomplished vary from vendor to vendor, although a flat-field acquisition at two or more exposure levels is generally required. Changes in the imaging geometry, beam energy, or image field between calibration and image acquisition can result in image artifacts. In the example in Figure 5, a flat-field correction was acquired and applied; however, an artifact appeared during routine quality control testing. The artifact in Figure 5 was traced to a defect in the aluminum x-ray beam filter. For a fixed relationship between the x-ray beam and detector, it would be anticipated that the detector calibration would have removed this artifact. However, because subsequent images were obtained with a different source-to-image distance than the detector calibration, the projection of the defect was shifted, resulting in the artifact. To resolve Fig. 6 Flawed gain calibration., Radiograph of frontal humerus shows appearance of phototimer sensors in patient image (arrows). B, Flat-field image clearly shows phototimer sensors. this artifact, the beam filtration was replaced and the detector was recalibrated. The images in Figure 6 represent another imaging issue related to calibration conditions. The images are from a digital detector that is used in a Bucky table. Manufacturer recommendations specify removing the detector from the Bucky table for calibration. Subsequent flatfield images clearly show the phototimer sensors used for automatic exposure control that are present in the Bucky table. The phototimer sensors are most visible in areas of unattenuated radiation and to a small extent across the skin line of the extended arm in Figure 6. Technically, this might not be considered an artifact because the image is showing exactly what is present in the beam under the specified operating conditions. However, with the use of DR (instead of CR) it should be possible to remove this to provide a better clinical image. Subsequent calibrations with the detector inside the Bucky table removed the appearance of the sensors. There are both advantages and disadvantages to the gain calibration available with B DR systems. On the one hand, it is necessary to compensate for the intrinsic variation in detector response. On the other hand, the gain calibration process could also lead to artifacts being written into subsequent images. If the acquisition parameters do not match those of a calibration (e.g., different source to image distance, different kilovoltage, and different detector position relative to the x-ray beam), an image is vulnerable to the appearance of artifacts. It is therefore advisable to calibrate detectors using clinical acquisition parameters. Vendors might also provide a means to have multiple calibration files available for different clinical settings (e.g., detector with and without a Bucky table). Furthermore, it is important to ensure that the calibration is successfully applied before resuming patient imaging by following any calibration with a visual inspection of a flat-field image. n example of a poor gain calibration is shown in Figure 7. In this image, the gain calibration was not applied, resulting in lines that correspond with individual tiles within the detector panel. Detector Lag and Flawed Offset Correction It is also possible to produce artifacts that appear as a signal deficit instead of excess residual signal, shown in Figures 3 and 4. These may result from a poorly timed offset (dark-noise) correction acquired while there is still residual lag signal from a previous acquisition. Offset correction determines the amount of detector signal present without additional exposure to the detector. Some vendors perform this correction on a regular basis (e.g., daily) or in response to a detector event, signal, or monitored environmental conditions. Figure 8 shows a series of images taken during a single patient examination. The first acquisition includes a lead marker in an area of unattenuated radiation. Subse- B JR:198, January

5 Walz-Flannigan et al. Fig. 7 Flawed gain calibration. Frontal chest radiograph shows artifact caused by poor gain correction. quent images contain artifacts in the detector location that appear as inverse lag signals of the lead marker. The inverse lag signal grows stronger over time and is still present up to 10 minutes later, as shown in Figure 8E. The significant unattenuated radiation in the first acquisition would have resulted in image lag that would look similar to the original image in that it would be radiopaque in appearance. The inverse signal artifact seen in subsequent images was caused by an offset correction that occurred shortly after the initial image acquisition. The inverse of the lag signal was written into the detector correction, thereby producing an artifact that appeared as a radiolucent marker. The contrast of the radiolucent artifact increases in time as the lag from the initial image decays away. These artifacts can have significant clinical consequence by creating ambiguous or mistaken laterality markers that obscure or confound findings with the inverse signal of anatomy from previous images. In contrast to lag artifacts, these artifacts do not decay over time but instead become stronger after a lag signal has decayed away, remaining until a subsequent offset correction occurs. Vendors may implement more frequent offset corrections to compensate for image lag and reduce the amount of wait time needed between image acquisitions. However, this requires repeated and continual recorrection as the lag signal decays and creates vulnerability to the type of artifact just described. If lag is a concern, it is preferable to increase the time interval between certain acquisitions. Vendors could monitor the amount of lag signal present and prevent acquisitions while it is high, but this process could also cause long wait times between image acquisitions and reduce efficiency. Backscatter Wireless (or tethered) DR detectors can have many workflow advantages. They can permit a wider variety of examinations than a fixed-wall or table DR detector, thus bringing DR to portable imaging or eliminating the need for additional CR views in fixed radiographic rooms. Wireless detectors are handled by a technologist the same way as a CR cassette. Thus, ergonomics and durability are important considerations. Heavier detectors would not be a favorable option for technologists and would increase the likelihood of the detector being dropped or mishandled. However, efforts to minimize detector weight could mean compromising shielding on the backside of the detector. Reduced shielding, challenging portable imaging configurations, and high exposures may create conditions for generation of images that are contaminated by backscattered radiation. rtifacts caused by backscatter are illustrated in Figure 9. These portable radiographs were acquired with higher exposures needed for a larger patient, thus creating a significant amount of scattered radiation. lack of appropriate collimation of the x-ray beam to the edges of the detector also results in increased scatter radiation. Scattered radiation incident upon the back of the detector can create an artifact by producing an image of the electronic components of the detector overlaid on the patient image. projection image of the DR detector alone clearly identifies the source of the artifacts in clinical images. To reduce the appearance of these Fig. 8 Detector lag with flawed dark-noise (offset) correction., Proximal femur radiograph shows lead markers at left. B E, Subsequent radiographs show inverse of lag image from (boxes). Images are taken 2 minutes 45 seconds (B), 3 minutes 7 seconds (C), 6 minutes 59 seconds (D), and 11 minutes 49 seconds (E) after acquisition of proximal femur image in. Note additional appearance of inverse of skin line from proximal femur radiograph visible in D and E indicated by arrows. 160 JR:198, January 2012

6 rtifacts in Digital Radiography artifacts, extra lead backing was attached to the detector to protect it from backscatter. Discussion DR is susceptible to many of the same artifacts as CR, including grid-related artifacts or those related to image processing. These types of artifacts are not included in this article because they have been well-addressed elsewhere [1 6]. This article includes other examples of well-recognized DR-specific artifacts (e.g., detector lag, faulty gain calibration) and varieties of artifact that have not yet been published in the form shown here (e.g., image stitching, backscatter). In addition to the artifact examples, we have provided strategies for avoiding or addressing these artifacts. Even though different systems may possess similar hardware, they may be more or less susceptible to image artifacts depending on how the vendor handles features such as timing of exposures, dark-noise corrections, gain calibration, image processing, or detector shielding. Some artifacts can cause patient safety concerns (e.g., ambiguous or mislabeled laterality due to lag); others could obscure or alter the appearance of diagnostically relevant anatomy. With new imaging equipment models, it is wise to be a wary consumer and vigilant for artifacts resulting from problematic implementations or unintended consequences. Fig. 9 Backscatter artifact. and B, Radiographs of lateral abdomen obtained with portable digital radiography detector show components of detector resulting from backscatter reaching detector through electronics at detector back. C, Image shows wireless digital radiography detector obtained with computed radiography cassette. To ensure good patient care it is important that technologists, physicists, and radiologists understand the physical limitations of flat-panel detectors and use equipment in ways that do not induce artifacts. s with all good practice in radiology, incorporating these image acquisition units into a quality control program is essential. Several guidelines for quality control programs for digital detectors have been published [10, 12]. Just as important is a familiarity with the visual appearance of these artifacts so they can be distinguished as artifact and the sources addressed. References 1. Solomon SL, Jost RG, Glazer HS, Sagel SS, nderson DJ, Molina PL. rtifacts in computed radiography. JR 1991; 157: Volpe JP, Storto ML, ndriole KP, Gamsu G. rtifacts in chest radiographs with a third-generation computed radiography system. JR 1996; 166: Cesar LJ, Schueler B, Zink FE, Daly TR, Taubel JP, Jorgenson LL. rtefacts found in computed radiography. Br J Radiol 2001; 74: merican ssociation of Physicists in Medicine. cceptance testing and quality control of photostimulable storage phosphor imaging systems: report 93. merican ssociation of Physicists in Medicine Website. RPT_93.pdf. Published October ccessed September 12, 2011 B 5. Shetty CM, Barthur, Kambadakone, Narajynan N, Kv R. Computed radiography image artifacts revisited. JR 2011; 196:157; [web]w37 W47 6. Honey ID, MacKenzie. rtifacts found during quality assurance testing of computed radiography and digital radiography detectors. J Digit Imaging 2009; 22: Siewerdsen JH, Jaffray D. ghost story: spatiotemporal response characteristics of an indirectdetection flat-panel imager. Med Phys 1999; 26: Willis CE, Thompson SK, Shepard JS. rtifacts and misadventures in digital radiography. ppl Radiol 2004; 33: Yorkston J. Flat-panel DR detectors for radiography and fluoroscopy. In: Goldman LW, Yester MV, eds. Specifications, performance evaluations, and quality assurance of radiographic and fluoroscopic systems in the digital era. Madison, WI: Medical Physics Publishing, 2004: Institute of Physics and Engineering in Medicine. Report 91:recommended standards for routine performance testing of diagnostic x-ray imaging systems. York, England: Institute of Physics and Engineering in Medicine, Machida H, Yuhara T, Mori T, Ueno E, Moribe Y, Sabol JM. Optimizing parameters for flat-panel detector digital tomosynthesis. RadioGraphics 2010; 30: Marshall NW, Mackenzie, Honey ID. Quality control measurements for digital x-ray detectors. Phys Med Biol 2011; 56: C FOR YOUR INFORMTION This article is available for CME credit. See for more information. JR:198, January