X-Ray-Based Medical Imaging and Resolution

Residents Section Physics Minimodule Huda and Abrahams Resolution on Radiographs Residents Section Physics Minimodule Residents inradiology Walter Huda 1 R. Brad Abrahams 2 Huda W, Abrahams RB Keywords: focal blur, image quality, motion blur, radiographic imaging, receptor blur, resolution DOI:10.2214/AJR.14.13126 Received May 9, 2014; accepted after revision May 30, 2014. 1 Department of Radiology and Radiological Science, Medical University of South Carolina, 96 Jonathan Lucas St, MSC 23, Charleston, SC 29425-3230. Address correspondence to W. Huda (walterhuda@hotmail.com). 2 Department of Radiology and Imaging, Georgia Regents University, Augusta GA. WEB This is a web exclusive article. AJR 2015; 204:W393 W397 0361 803X/15/2044 W393 American Roentgen Ray Society X-Ray-Based Medical Imaging and Resolution T he purpose of most radiologic imaging examinations is to either identify abnormalities in a patient or to classify a patient as being healthy while minimizing radiation exposure [1]. To achieve these goals, it is of obvious importance to ensure that the radiologic image quality is sufficient for a given imaging task. Medical image quality is normally characterized in terms of contrast, noise, and spatial resolution. The achieved quality for any image will depend on both the intrinsic properties of the imaging system and the manner in which the images were obtained, including the choices of image acquisition tube voltage (kilovoltage) and output (tube current exposure time product). To be detectable, a lesion must transmit a different x-ray beam intensity compared to the surrounding normal tissues. The difference in x-ray beam intensity through a lesion relative to the normal background is called subject contrast. The appearance of this subject contrast in the resultant image is called image contrast. Noise has both random and structural components, with the latter being fixed for a given patient, such as ribs in chest radiographs. The random noise in virtually all radiographic imaging is due to quantum mottle, which can be reduced by using more x-ray photons to create the image, but which will also increase the patient dose [2]. Contrast and noise are directly related to the choice of radiographic techniques, namely x-ray tube voltage (kilovoltage) and output (tube current exposure time product) [3]. The amount of blur, or unsharpness, affects the appearance of both normal anatomy and pathologic abnormalities. Spatial resolution is the technical term used to refer to the amount of blur in an image. Spatial resolution performance is an intrinsic property of an imaging system that is generally independent of the selected technique factors (kilovoltage and tube current exposure time product). In this article, we describe the essential characteristics of spatial resolution. Factors that influence spatial resolution, as well as performance of x-ray-based medical imaging modalities, are described [4 7]. What Is Resolution? Blur and Sharpness Most radiologists have an intuitive sense of what constitutes poor resolution and can easily recognize a blurred image. Resolution, however, is sometimes taken to mean the ability to see objects so that when lesions are seen, then the resolution must be good. Such statements are incorrect, as illustrated by considering a single-cell molecule labeled with a radionuclide and imaged with a PET system. The activity will be seen, but the blur associated with this point source will be several millimeters, resulting in a blurry image with poor resolution. This example can be expanded to illustrate what imaging scientists mean regarding spatial resolution performance. For two nearby cells with radioactively labeled DNA in their nuclei, a system that can resolve these two entities would show two hot spots as distinct entities (Fig. 1, upper panel). A PET system would show only one very large blurred object, and the activity in cell one would not be resolved from the activity depicted in the nearby cell two (Fig. 1, lower panel). A system with good resolution permits the detection (and characterization) of two small features, such as microcalcifications in mammograms, even when they are physically close to each other. There are several terms used to describe spatial resolution in imaging. Blur and sharpness are good descriptors that are universally understood, and they simply mean that a sharp edge will also appear as sharp (not blurred) in an image obtained with a system W393

Huda and Abrahams Fig. 1 Objects are depicted on left, and resultant images are shown on right. Top panel shows that two objects that are separated far apart result in two (blurred) lesions in image. Bottom panel shows that, when these two objects move closer together, image shows one blurry object, and lesions are not resolved. that has good resolution performance. Other terms that may be used include resolution, high-contrast resolution, unsharpness, and detail visibility. Measuring Resolution Imaging scientists consider the spatial resolution performance to be an intrinsic characteristic property of an imaging system. In general, resolution cannot be evaluated by looking at line pair phantoms because the visibility (detectability) in phantoms depends on the amount of noise in the image. It is only when noise is somehow made negligible that one is left with the intrinsic blur associated with a given imaging system. Two common ways of assessing the intrinsic resolution properties of an imaging system are the line spread function and the modulation transfer function. A line spread function is measured using an image obtained of a narrow line, which is simply a blurred version of the initial line (Fig. 2). The amount of blur can be measured as the full width half maximum, which for a line is the physical distance in millimeters between the two points where the line intensity has been reduce to half of the central (maximum) value. Different measures of the spread in the image are also possible, such as the full width tenth maximum, where the value of full width tenth maximum is obviously greater than the full width half maximum (Fig. 2). The advantage of these measures is that they are intuitive and provide a quantitative sense of the amount of blur in a given imaging system. A modulation transfer function curve can be estimated by looking at the images obtained at differing spatial frequencies, as depicted in Figure 3. For low spatial frequencies, corresponding to very large bars and gaps, the region behind the lead will be pure white and the region in the gap between the bars will be pure black. The difference between the lead and gap regions is called modulation and is essentially close to 100% (black vs white) at the lowest spatial frequencies. This modulation is always reduced as spatial frequencies increase because of the blurring effects of an imaging system. At some high Fig. 2 Illustration on left depicts radiographic image being generated of narrow line of x-rays, such as might result when x-ray passes through long narrow slit in sheet of lead. Upper right panel shows broadened (blurred) line in radiographic image; lower right panel shows line profile through radiographic image, showing line spread function (LSF) and measured broadening as full width half maximum (FWHM) and full width tenth maximum (FWTM). (Illustration by Abrahams RB) spatial frequency, the image of the bars and gaps blends into a uniform gray. At this point, there is zero modulation in the image, and adjacent bars cannot be differentiated (Fig. 3). Line Pair Phantoms and Pixel Size A line pair phantom contains strips of lead, with gaps of the same size as the lead. When the lead (and gap) are 0.5 mm, we can fit one line pair into 1 mm and call this a spatial frequency of one line pair per millimeter. High spatial frequencies correspond to small objects and vice versa. When a line pair phantom is imaged, the large objects are readily resolved from each other because there is a large gap between adjacent lines (Fig. 3). As the spatial frequency increases, however, the lines start to blur into each other and the limiting spatial resolution occurs when distinct lines are no longer visible (Fig. 3). When the images are obtained at high exposures so that noise is negligible, this spatial frequency is known as the limiting spatial frequency and is expressed in line pairs per millimeter. The use of large pixels would clearly reduce the limiting spatial resolution performance. In general, however, the use of small- Fig. 3 On left is bar pattern that ranges from low spatial frequencies (largest objects) to high spatial frequencies (smallest objects), but all have 100% modulation defined as difference in intensity (black vs white). On right is resultant image showing how image blur reduces modulation from close to 100% for large objects to 0% modulation at spatial frequency corresponding to limiting spatial resolution. W394

Resolution on Radiographs er pixels is not expected to improve spatial resolution performance. The reason for this is that there are intrinsic sources of blur in the imaging chain, such as the size of the focal spot, that cannot be overcome by making pixels smaller. Vendors of imaging systems will make the pixels as small as they need to be, and pixel size is not further reduced when intrinsic factors become the determinant factors of achievable image sharpness. What Affects Resolution? Focal Spot The size of the focal spot affects the amount of focal spot blur in all radiologic imaging, as depicted in Figure 4 [8 10]. In radiography, fluoroscopy, and CT, the x-ray tube normally has two focal spot sizes, with nominal sizes of 0.6 and 1.2 mm. When the large focal spot is used, the total power loading that may be used is 100 kw, whereas the small focal spot can only tolerate 25 kw. The use of small focal spots will thus generally result in longer exposures, which, in turn, increase the likelihood of motion blur (see discussion later in this article). In mammography, the x-ray tube normally also has two focal spot sizes, with nominal sizes of 0.1 and 0.3 mm. The maximum tube current with the small focal spot is only 25 ma but can be four times higher with the large focal spot. As with radiography, the use of a small focal spot is generally associated with much longer exposures and the associated problems of motion blur. In mammography, geometric magnification is used to investigate suspicious regions, which is achieved by moving the compressed breast closer to the x-ray tube (Fig. 4). In interventional neuroradiology, the visibility of very small blood vessels can be improved through the use of geometric magnification. In both of these cases, the use of geometric magnification will always require the use of smaller focal spots to minimize focal spot blur. Motion The amount of motion blur is directly related to the speed of any motion, which may Fig. 4 Relationship between size of focal spot and amount of blur in resultant image in magnification mammography. Use of a large focal spot (top) results in a much blurrier image than small focal spot (bottom). This is especially true for magnified images depicted in this graphic, illustrating importance of small focal spot when geometric magnification (twofold) is being used. (Illustration by Abrahams RB) be voluntary or involuntary. Examples of involuntary motion include cardiac motion and peristalsis. The amount of motion blur is also directly proportional to the exposure time of any radiographic examination. Motion blur can be estimated by multiplication of the speed of any motion by the corresponding exposure time. When increasing x-ray tube output, the x-ray tube currents should be increased when this is technically feasible, so that exposure times are kept as short as possible. Chest radiographic imaging has very short exposure times, of the order of milliseconds, which is easy to achieve as the chest is not difficult to penetrate and only low x-ray tube outputs ( 1 mas) are used. In abdominal imaging, exposure times must be increased to tens of milliseconds to ensure that enough radiation is incident on the image receptor, which is achieved by use of much higher intensities ( 20 mas). The longest exposures are generally encountered in mammography, where a typical exposure time is approximately 500 ms or more and up to a factor of 3 higher when performing magnification mammography [11]. There are a number of practical steps that may be taken to minimize patient motion. Immobilization devices include head supports in CT and compression paddles in mammography [12, 13]. Patients can be asked to hold their breath and stay still. In cardiac CT, β-blockers can be prescribed to reduce the heart rate [14]. Sedation can be used when imaging newborns and young infants [15]. Receptor Size Scintillators used in radiology include cesium iodide used in flat-panel detectors and image intensifiers and gadolinium oxysulfide in mammographic screens. When x-rays are absorbed by any scintillator, approximately 10% of the absorbed energy is converted into light, which spreads out when traveling to the light detector, as depicted in Figure 5 [16]. This spreading of light between the x- ray interaction and the corresponding light detector (film or a digital equivalent) may result in significant image blur. Although a columnlike structure can be used to reduce diffusion of light, detector blur is nonetheless present in all scintillators. The amount of blur is directly related to the scintillator thickness as depicted in Figure 5. In photoconductors, charge created after a photoelectric or Compton interaction is collected directly by the application of an electric field [17]. Because low-energy electrons do not travel very far, there will be very little blur (Fig. 6). Consequently, photoconductors are expected to have excellent resolution performance and are attractive for use in digital mammography, where good resolution performance is Fig. 5 For scintillators that absorb x-rays and produce light, amount of image blur is directly related to thickness of scintillator (1 mm [left] vs 0.5 mm [right]). Thickness of scintillator can be taken to be comparable to blur introduced into resultant image. W395

Huda and Abrahams Fig. 6 Illustration of photoconductor shows that charge (electrons) generated by incident photons have minimal divergence when collected by application of voltage across photoconductor. essential. By contrast, in computed radiography, light beams are used to release the energy stored in an exposed photostimulable phosphor. As shown in Figure 7, there will be scattering of the incident light, which will result in increased image blur because of the release of light from adjacent pixels. For this reason, there is a limit to the thickness of any photostimulable phosphor. When thin detectors are used, this increases patient doses because of the reduction in x-ray absorption in thinner detectors. In fluoroscopy, the number of television lines determines the nominal width of each line and thereby affects the achievable spatial resolution performance. When the number of television lines increases from 500 to 1000, the limiting resolution performance will also improve by a factor of 2. In CT, it is the physical size of each detector (length and width) that is directly related to the achievable spatial resolution performance [18]. When corrected for image magnification, the nominal detector dimension in both directions on a CT scanner may be taken to be a nominal 0.5 mm or so. Fig. 7 Photostimulable phosphor shows scattering of incident read-out narrow light beam (left) and emission of stimulated light from pixel of interest (right), as well as adjacent pixels, which increase image blur. Resolution Performance in Radiographic Imaging Radiography and Mammography Digital radiographs generally have a matrix size of 2000 2500. For a 35 43-cm cassette, this digital matrix corresponds to a pixel size of 175 μm and a limiting resolution of three line pairs per millimeter. For a 20 24-cm cassette, the pixel size would be 100 μm, and the limiting resolution would be five line pairs per millimeter. Extremity radiographs, where good resolution is important for detecting hairline fractures, should therefore be obtained using smaller cassettes. The limiting resolution of a typical film-screen system used for chest radiography was about six line pairs per millimeter, or a factor of 2 better than a typical digital chest radiographic imaging system. In digital mammography, the pixel size typically ranges between 50 and 100 μm, so that the corresponding limiting spatial resolution ranges between 10 and five line pairs per millimeter. A common pixel size of 70 μm corresponds to a limiting resolution of close to seven line pairs per millimeter. The total number of pixels in a digital mammogram is about 12 million, assuming a 3000 4000 matrix size. Most mammography workstations are capable of displaying up to 5 million pixels, so that seeing all the available detail in a digital mammogram will require the use of image zoom capabilities [19]. Film-screen mammography typically achieved a limiting spatial resolution of 15 line pairs per millimeter, which is a factor of 2 better than digital mammography. This example illustrates that the benefits of digital mammography relate to image processing rather than any resolution issue per se. Mammography film-screen systems are superior to the capabilities of the human visual system, which is about five line pairs per millimeter at a 25-cm viewing distance [19]. This explains why mammographers in the days of film always had a magnifying glass in their hands for viewing mammograms on viewboxes. Fluoroscopy and Digital Photospot Fluoroscopy can be performed with image intensifiers or flat-panel detectors and does not generally require high-resolution performance. With a standard 500-line television system, the nominal limiting fluoroscopy spatial resolution performance is about one line pair per millimeter. Flat-panel detectors with a pixel size of 175 μm can achieve a limiting resolution of three line pairs per millimeter. When these are used for performing fluoroscopy, binning four pixels into one large pixel is common for larger FOVs, which results in a limiting resolution of 1.5 line pairs per millimeter. Digital photospot images using an image intensifier based imaging chain are generally obtained using a high-quality (1000 line) television system. For a 25-cm FOV, the resultant pixel size is 250 μm, and the corresponding limiting resolution will be about two line pairs per millimeter. Commercial 2000-line television systems are available, but the benefits of providing improved spatial resolution performance were not deemed to be worthwhile. For flat-panel detectors, spatial resolution for digital photospot images is identical to that for radiographs obtained with these detectors namely, three line pairs per millimeter. CT In a head CT scan, the FOV is 250 mm, the matrix size is 512 512 and the pixel size is 0.5 mm. The best achievable resolution for this FOV and matrix size is thus one line pair per millimeter. Increasing the FOV to 500 mm for a large patient doubles the pixel size (1 mm) and halves the limiting spatial resolution performance (i.e., 0.5 line pair per millimeter). Spatial resolution in CT is thus an order of magnitude worse than filmscreen combination and four times worse W396

Resolution on Radiographs TABLE 1: Representative Values of Limiting Resolution Performances for Digital Imaging Systems Imaging System Limiting Resolution, Line Pairs per Millimeter Comments CT 0.7 Improved by zoom reconstruction of the central region of acquired image Fluoroscopy 1 Improved by the use of electron magnification modes, as well as high-quality television systems (1000 line) Digital subtraction angiography or photospot 2 Improved by the use of electronic magnification modes of image intensifiers (e.g., magnification 1 or 2) Radiography 3 Improved performance can be achieved by the use of smaller cassette sizes (5 line pairs per millimeter for 20 25 cm) Mammography 7 Use of smaller (50 μm) pixels would improve resolution, whereas large pixels (100 μm) would degrade resolution than digital chest radiography. CT has excellent imaging characteristics and anatomic localization compared with any type of projection imaging but also has inferior spatial resolution performance. In CT, the choice of the mathematic reconstruction filter affects the amount of blur in the resultant image [20]. Image reconstruction filters offer reduced image noise but at the expense of more image blur. Common names for such high-resolution reconstruction filters include bone, detail, high resolution, and lung. These filters are used to get excellent detail for structures such as bone (vs air or tissue) or lung (vs air), where the high intrinsic contrast negates the importance of higher levels of noise. Reconstruction filters with names such as standard or soft tissue reduce image noise at the price of inferior spatial resolution performance and are used where the intrinsic lesion contrast is low. CT spatial resolution can be improved by up to a factor of about 2 by reducing the reconstruction FOV (zoom). This requires that the original projection data are used to reconstruct a smaller anatomic region, which thereby differs from simple magnification, which uses reconstructed image data. Obtaining the best possible resolution may also require the use of a reduced focal spot (0.6 vs 1.2 mm), which will also require a reduction of the focal spot power loading (25 vs 100 kw). However, when a small focal spot is used with reduced power loading, the scan time may need to be increased, which also increases the likelihood of motion blur. Conclusion Resolution is the ability of an imaging system to faithfully reproduce a sharp edge that is present in the object. The key factors that influence the sharpness of an image relate to the size of the source of x-rays (focal spot), the physical characteristics of the x-ray detector system (area and thickness), and the presence of any motion blur because of the finite duration of all radiographic exposures. Image processing, especially in CT through the filtered back projection reconstruction, can also affect spatial resolution performance. The limiting resolution that is routinely achieved in clinical practice ranges from about 0.7 line pair per millimeter for CT to about seven line pairs per millimeter for digital mammography (Table 1). The values listed in Table 1 may be compared to the human visual system, which can resolve up to five line pairs per millimeter at a normal viewing distance of 25 cm. References 1. Huda W. Understanding (and explaining) imaging performance metrics. AJR 2014; 203:[web] W1 W2 2. Huda W. Kerma area product in diagnostic radiology. AJR 2014; 203:[web]W565 W569 3. Huda W, Abrahams RA. Radiographic techniques, contrast, and noise in x-ray imaging. AJR 2015; 204:W126 W131 4. Bushberg JT, Seibert JA, Leidholdt EM, Boone JM. The essential physics of medical imaging, 3rd ed. Philadelphia, PA: Lippincott, Williams & Wilkins, 2012 5. Huda W. Review of radiologic physics, 3rd ed. Philadelphia, PA: Lippincott Williams & Wilkins, 2010 6. Nickoloff EL. Radiology review: radiological physics. Philadelphia, PA: Elsevier Saunders, 2005 7. Wolbarst AB. Physics of radiology, 2nd ed. Madison, WI: Medical Physics Publishing, 2005 8. Schueler BA. Clinical applications of basic x-ray physics principles. RadioGraphics 1998; 18:731 744; quiz, 729 9. Villafana T. Generators, x-ray tubes, and exposure geometry in mammography. RadioGraphics 1990; 10:539 554 10. Zink FE. X-ray tubes. RadioGraphics 1997; 17:1259 1268 11. Hogge JP, Palmer CH, Muller CC, et al. Quality assurance in mammography: artifact analysis. RadioGraphics 1999; 19:503 522 12. Ayyala RS, Chorlton M, Behrman RH, Kornguth PJ, Slanetz PJ. Digital mammographic artifacts on full-field systems: what are they and how do I fix them? RadioGraphics 2008; 28:1999 2008 13. Barrett JF, Keat N. Artifacts in CT: recognition and avoidance. RadioGraphics 2004; 24:1679 1691 14. Pannu HK, Alvarez W Jr, Fishman EK. Betablockers for cardiac CT: a primer for the radiologist. AJR 2006; 186(suppl 2):S341 S345 15. Macias CG, Chumpitazi CE. Sedation and anesthesia for CT: emerging issues for providing highquality care. Pediatr Radiol 2011; 41(suppl 2):517 522 16. Haus AG. The AAPM/RSNA physics tutorial for residents: measures of screen-film performance. RadioGraphics 1996; 16:1165 1181 17. Rowlands JA, Zhao W, Blevis IM, Waechter DF, Huang Z. Flat-panel digital radiology with amorphous selenium and active-matrix readout. Radio- Graphics 1997; 17:753 760 18. Sprawls P. AAPM tutorial: CT image detail and noise. RadioGraphics 1992; 12:1041 1046 19. Samei E. AAPM/RSNA physics tutorial for residents: technological and psychophysical considerations for digital mammographic displays. Radio- Graphics 2005; 25:491 501 20. Goldman LW. Principles of CT: radiation dose and image quality. J Nucl Med Technol 2007; 35:213 225; quiz, 226 218 W397