Super-Resolution Applications in Digital Radiography
 
Authors:
Leonard F. Berliner, MD, New York Methodist Hospital
 
Hypothesis:
Techniques of super-resolution (SR), when applied to digital radiography, may be utilized to achieve images of diagnostic quality (SR-Radiographs). Techniques may be employed to obtain: (a) low-dose or variable-dose radiographic images, (b) high-resolution images that can exceed the intrinsic spatial resolution of the detector, and (c) high-resolution angiography (SR-Angiography) that exceeds the spatial resolution of currently available technology. These techniques have the potential for substantially lowering radiation exposure to patients, such as neonates and pediatric patients, and for providing high-resolution digital radiography and angiography.
 
Introduction:
Ionizing radiation from medical imaging now represents over 95% of all man-made radiation exposures, and is the single largest radiation source after natural background radiation.(1) Techniques of super-resolution, when applied to digital radiography, can achieve a low-dose alternative to conventional digital radiography.(2)(3) Super-resolution radiography (SR-Radiography) may be described as image processing that transforms a sequence of standard (low resolution and low dose) fluoroscopic frames, into an enhanced, high resolution radiographic image. This process, when applied to a series of fluoroscopic frames, can also produce a high resolution angiographic sequence.

Super-resolution, as described in optical engineering, is a process in which there is fusion of a sequence of low-resolution, noisy and/or blurred images (often photographic or video) to produce a higher resolution image or sequence. In addition to increasing resolution, when the individual frames are in register, the process of averaging the individual frames also results in improved signal-to-noise.

Through the application of this technology in digital radiography, super-resolution techniques may be applied in a variety of settings to achieve low-dose and high resolution images, such as neonatal and pediatric imaging and angiography.
 
Methods:
 
Figure 1

Figure 1 The super-resolution radiograph (SR-Radiograph) is processed
from a consecutive group of low-resolution, fluoroscopic, component frames.
 

The super-resolution digital radiography and angiography produced in this study utilized standard fluoroscopic sequences that were obtained with an OEC C-arm or GE LUA angiography suite utilizing conventional (1k x 1k) image intensifiers. The “component frames” that were utilized in this study were actually of 720 x 720 pixel resolution.

The processing for SR-Radiography required the following steps: (a) division of a portion of the original fluoroscopic sequence into 10 individual frames, (b) subpixel registration of the selected group of 10 frames, as well as the registration of intraframe components (which sets this process apart from frame-averaging), (c) combination of the registered frames, and (d) post-processing to adjust brightness, contrast, and other image parameters. (Figure 1)

The process for obtaining SR-Angiography requires the sequential creation of individual angiographic frames (using the technique described above). The individual SR-Angiographic frames (actually a series of high-resolution SR-Radiographs) are reassembled and viewed sequentially as an angiographic sequence.

SR-Radiographs and SR-Angiograms were compared to (a) the component images, (b) images obtained through simple frame-averaging, and (c) conventional digital radiographs obtained at 2000 x 2000 pixel resolution (2K x 2K) from commercial Swissray digital radiographic equipment. Evaluation of images of a standard radiographic line-pair resolution phantom was performed.

A technique for simulating low-dose, or variable-dose, SR-Radiography was devised. In certain clinical settings, fewer than 10 frames may be selected, thereby reducing radiation dose, and still achieving a diagnostic image. Full-dose, SR-Radiographs (composed of 10 component frames) were compared to half-dose, SR-Radiographs (composed of 5 component frames).
 
Results:
The ability to improve upon spatial resolution was tested with a radiographic line pair resolution phantom. Figure 2 presents a comparison of a single fluoroscopic frame (Figure 2A) with a SR-Radiograph (Figure 2B) constructed from 10 component frames. When viewed at full size, the single fluoroscopic frame displays a resolution approaching 2.5 line pairs per mm, while the SR-Radiograph displays a resolution approaching 3.0 line pairs per mm. The line-pair resolution achieved with SR (using 720 x 720 component images) exceeds line-pair resolution of standard (2K x 2K) digital radiography (Figure 2C) (Image: ICRP)(4).
 
Figure 2

Figure 2 The SR-Radiograph (Figure 2B), constructed from 10 component frames, is compared with a single fluoroscopic frame (Figure 2A) and a standard (2K x 2K) digital radiograph (Figure 2C) (Image source: ICRP) (Reference 4). The SR-Radiograph displays a resolution approaching 3.0 line pairs per mm, while the single fluoroscopic frame and the standard digital radiograph display a resolution approaching 2.5 line pairs per mm.
 
SR-Radiography image quality is demonstrated in Figure 3. A single component frame (Figure 3A) may be compared with a SR-Radiograph (Figure 3B) obtained from 10 component frames. Side-by-side comparison demonstrated that SR-Radiography produced images which were comparable to conventional 2K x 2K digital radiographs. (Figure 3C)
 
Figure 3

Figure 3 A single component frame (Figure 3A) is compared with a SR-Radiograph (Figure 3B) obtained from 10 component frames. The image quality of the SR-Radiograph may be compared with similar view from a standard ( 2K x 2K) digital radiograph (Figure 3C). (Image source: Swissray)
 
Figure 3 A single component frame (Figure 3A) is compared with a SR-Radiograph (Figure 3B) obtained from 10 component frames. The image quality of the SR-Radiograph may be compared with similar view from a standard ( 2K x 2K) digital radiograph (Figure 3C). (Image source: Swissray)
 
Figure 4

Figure 4 A single unsubtracted (Figure 4A) and subtracted (Figure 4C) component frame is compared with an unsubtracted (Figure 4B) and subtracted (Figure 4D) SR-Angiogram which was obtained from 10 component frames. (The component frame and SR-Angiogram are matched with respect to the time frame of the angiographic sequence.)
 
 
 
 
Figure 4 demonstrates the image quality obtained with SR-Angiography. A single component frame (Figure 4A and 4C) is compared with a SR-Angiogram (Figure 4B and 4D), which was obtained from 10 component frames.
 

The improvement in image quality in SR-Radiography over frame-averaging was demonstrated. Each image was constructed from the same set of 10 component frames. However, increased detail was achieved through super-resolution.

Variable-dose radiography was simulated. Figure 5 demonstrates that the half-dose image is not simply under-exposed, but contains considerable detail. Although there is an increase in noise when compared to the full-dose image, in certain clinical situations, the reduced dose image may provide adequate diagnostic information.
 
Discussion:
Research on super-resolution in recent years verifies and supports the claims of increased resolution, even beyond the Nyquist resolution of an imaging system, as well as noise reduction. In the most common super-resolution algorithms, the information gained in the super-resolution image is embedded in the low resolution, under-sampled images in the form of aliasing. Substantial detail can be extracted from the component low-resolution frames, provided that there is sufficient subpixel overlap frame to frame.(5) Under practical conditions, the limit of increase in resolution has been found to be 1.6. Under synthetic conditions, the theoretical limit has been calculated to be 5.7.(6) In addition to increasing resolution, when the individual frames are in register, the process of averaging the individual frames also results in improved signal-to-noise, which equals the square root of the number of frames combined.(7) For example, to achieve a signal-to-noise ratio that is four times better than a single component image, 16 images need to be summed.

Application of super-resolution in digital radiography has been demonstrated in this study. The results show that improvements in resolution and signal-to-noise can be achieved. Objective evidence is provided by studies of a line-pair resolution phantom. Subjectively, the SR-
Radiographs are on a par with conventional digital radiography, and the SR-Angiographic frames exceed the image quality of the component frames.

The technique of variable-dose SR-Radiography may provide Radiologists control over radiographic exposure. Selection of radiation exposure may be adjusted by the user to achieve an optimal image at an acceptable signal-to-noise ratio. This may be especially valuable in pediatric applications. For certain applications (such as a neonatal chest X-ray for central line placement where maximum image quality may not be required every time) it may be possible to reduce the number of fluoroscopic frames acquired, reduce radiation exposure, and still achieve adequate image quality. The Radiologist can strike a balance between radiation dose and image quality.
 

Figure 5

Figure 5 The technique to achieve variable-dose radiography was simulated. A single component frame is shown in Figure 5A (inverted for comparison). Figure 5B is a half-dose image which is not simply under-exposed, but contains considerable detail. Although noise is increased compared to the full-dose image (Figure 5C), in certain clinical situations, the reduced dose image may provide adequate diagnostic information.
 
Conclusion:
Super-resolution techniques applied to digital radiography is a promising technology. Preliminary results with component images obtained at 720 x 720 pixel resolution confirm that SR-Radiography, SR-Angiography, and variable-dose SR-Radiography are viable technologies.

Further studies utilizing flat-panel detectors, which have higher intrinsic resolution, are planned. In addition, studies are planned to confirm that images of sufficient image quality (in specific clinical applications) can be achieved using low-dose, or variable-dose, SR-Radiography.
 
References:
1. Valentin J. Managing Patient Dose in Digital Radiology: Guest Editorial, Preface, Main Points, Glossary and Chapter 1. Ann ICRP. 2004;34:1-73, 2004.

2. Berliner LF. United States Patent 6,278,765 entitled Process for producing diagnostic quality x-ray images from a fluoroscopic sequence. August 21, 2001.

3. Bernhardt P. Lendl M. Deinzer F. New Technologies to Reduce Pediatric Radiation Doses. Pediatr Radiol. 2006;36(Suppl2):212–215, 2006.

4. Valentin J. Managing Patient Dose in Digital Radiology: Appendix. Ann ICRP. 2004;34:51-73.

5. Park SC, Park MK, Kang MG. Super-Resolution Image Reconstruction: A Technical Overview. IEEE Signal Processing Magazine. 2003;20:21-36.

6. Lin Z, Shum HY. Fundamental Limits of Reconstruction-Based Superresolution Algorithms under Local Translation. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2004;26:1-15.

7. Berry R, Burnell J. The Handbook of Astronomical Image Processing, Second edition. Willmann-Bell, IncRichmond, VA. 2006;40.