Australasian Physical & Engineering Sciences in Medicine Volume 27 Number 2, 2004
TECHNICAL REPORT
A CMOS image sensor method of focal spot size measurement* T. Tuchyna1 and D. Paix2 1
Department of Medical Technology and Physics, Sir Charles Gairdner Hospital, Perth, Australia School of Information & Electrical Engineering, University of South Australia, Mawson Lakes, Australia
2
Abstract A phosphor opto-coupled monochrome CMOS image sensor with a slit diaphragm was used to investigate focal spot characteristics. Images were captured during x-ray exposure with a triggered frame grabber and subsequently enhanced. Dimensions of the focal spot width (1.39mm) and length (1.92mm) were determined from the focal spot intensity profiles and their corresponding Full Width at Half Maxima (FWHM) in two orthogonal orientations. The CMOS image sensor measurements demonstrated differences in the measured width and length dimensions when compared to film measurements. The obtained nominal focal spot values however showed that image-sensor determined focal spot dimensions agreed with the direct film and film-screen methods when based on the AS/NZS defined nominal focal spot values. The CMOS image sensor tested appears to lack the measurement accuracy required for the measurement of small focal spot sizes due in part to its limited camera sensitivity.
lower noise, easy circuit integration and reduced cost. In a preliminary assessment of the suitability of a CMOS image sensor for the capture and analysis of x-ray images, we evaluated a technique for measurement of the focal spot size for a radiographic tube, using a CMOS image sensor and the slit test tool. The performance of this technique is characterised and compared to the direct film and film/screen methods of focal spot size measurement. The three basic methods for the measurement of the focal spot in a radiography tube employ the capture and measurement of a pinhole, slit or star resolution pattern images8. The slit method is the preferred choice of manufacturers and is endorsed in the Australian/New Zealand Standard (AS/NZS) 4274:1995 report9. The size and shape of the focal spot is critical to the viewing quality of the resultant image, where viewing quality is interpreted as visibility and resolution of small details. In general radiography the focal spot is usually small enough to have little effect on the clinical value of most radiographs. However the focal spot can significantly degrade the image sharpness where resolution is critical, i.e. mammography10 and magnification work11. Further, the determination of the dimensions of the focal spot is useful since it ensures that the tube is operating correctly at installation and is continuing to work correctly.
Key words
focal spot, radiography, CMOS image sensor, electronic imaging, quality assurance
Introduction Charge coupled devices (CCDs) and their variant complimentary metal oxide semiconductor (CMOS) devices are finding an increasing number of image capture roles, having many research and industrial applications1-5. Most small field radiological imaging systems use the proven and reliable CCD technology. CCD systems have found routine application in small field mammography for stereotactic localisation. Speller et. al.6 used a CCD based camera for the production of pinhole focal spot images, whereas recently Rong et. al.7 evaluated computed radiography and flat panel amorphous silicon plates in their ability to measure focal spot size. CCDs must however be fabricated in special CCD chip foundries, they are expensive and do not integrate easily with other circuit and image system components. The CMOS process of design and fabrication results in an imaging system with a greatly improved dynamic range, *Presented in part at the Engineering & Physical Sciences in Medicine Conference, Hobart, Australia, 15 - 19 November 1998 Corresponding author: T. Tuchyna, Department of Medical Technology & Physics, Sir Charles Gairdner Hospital, Perth, Hospital Avenue, Nedlands WA 6009, Australia Tel: (08) 9346 4277, Fax: (08) 9346 3466, Email:
[email protected] Received: 28 January 2004; Accepted: 24 June 2004 Copyright © 2004 ACPSEM/EA
Materials The Image Sensor Module consists of an image sensor chip (Model: VVL 1011C-001, VLSI Vision, Edinburgh, Scotland) mounted on a printed circuit board (PCB). The chip
63
Tuchyna et al
Australas. Phys. Eng. Sci. Med. Vol. 27, No 2, 2004
is essentially a photodiode array with each photodiode forming a part of a MOS transistor. The charge collected by each photodiode is read out through the switching of the transistor. The photodiode array consists of 312 x 287 pixels2 forming a field of view of 6.1 x 4.6 mm2 with a pixel size of 19.6 x 16.0 Pm. The photodiodes are read out one row at a time. A vertical register successively activates the row lines while a horizontal register controls the sequential read-out within each column line. The image sensor module requires only a 5V DC power input and produces a 1.0V p-p composite monochrome video output. In order to utilise the image sensor capabilities, a fluorescent phosphor was opto-coupled to the photodiode array using silicone optical grease to minimise reflections between the phosphor and photodiodes. A literature search12-16 along with a qualitative assessment of several commercially available phosphors suggested the use of gadolinium oxysulphide (Gd2O2S:Tb), the phosphor used in the construction of the KODAK LANEX Regular screens. A small piece of the actual Kodak Lanex regular screen (7 x 5 mm) with a stated phosphor density of approximately 67 mg/cm2 was used. During irradiation, light that is generated in the phosphor screen is converted to charge within the photodiode, causing a current to flow between the photodiode and an amplifier. The charge is proportional to the light reaching the photodiode during irradiation. A clock crystal resident on the PCB governs the row and column readout rates, as well as controlling the time over which charge can be collected. The minimum integration time for the image sensor was 40ms. In order to capture the image sensor’s monochrome video output signal a low cost PCI based frame grabber (Model: DT3155, Data Translation, Hants, England) was used with a Pentium 100MHZ, Personal Computer, (PC). The frame grabber requires both horizontal and vertical sync signals. The Sync circuitry provides a means of informing the frame grabber where in the video signal to expect a new frame acquisition. The DT3155 can accept an external trigger so that image acquisition can be synchronised with an event external to the PC. This was achieved by designing and constructing an electronic circuit where the initial pulse was taken from the ‘X-ray On’ panel light of the x-ray generator, thus matching frame capture to the x-ray 'switch on' period. The circuit produced a TTL trigger pulse with a variable time delay required for the initiation of the frame grabber. The delay was used so that high frequency noise signals during the kV build– up of the x-ray tube did not cause pre-triggering of the frame grabber. An Odel Polaire OT 1201, fully rectified, single phase x-ray generator was used (1973 model). The x-ray machine selectable mA settings were 50, 100, 200 and 300 mA. An indicator on the panel showed that the small focal spot was used at 50 and 100 mA, while the large focal spot was used at 200 and 300 mA. Due to the age of the x-ray machine, it was not possible to obtain manufacturer focal spot data for the CGR tube. Owing to the small detector area of the photodiode array, accurate alignment of the image sensor with the central axis of the beam is important. To enable correct
x A CMOS image sensor method
positioning of the image sensor to the central axis of the xray beam, an adjustable x-y bench jig using a microscope stage was made. This permitted reasonably accurate two dimensional movement in 1mm graduations. The slit was manufactured from two tungsten pieces 1.90 mm thick in order to fully attenuate the primary x-ray beam and machined to dimensions specified in the AS/NZS 4274:1995 report, see figure 1. The blade-shaped tungsten pieces were placed into a specially machined aluminium holder with a tightly fitting channel to accommodate the blades, allowing movement in one dimension only. A 10 Pm spacer of mica crystal, chosen for its hardness, was obtained by splitting a mica block. The two tungsten blades were then brought together and secured in place by screws, separated only by the spacer, which was subsequently removed. Focal Spot
0.014 1.90
8.4o
7.70
8.70
// 0.0025
Dimensions in millimetres Figure 1. Dimensions of the manufactured slit diaphragm based on the AS/NZS 4274: 1995 report.
Methods The need for a scaling factor The captured image was displayed on an NEC 16" monitor with a pixel pitch of 0.28 mm and capable of displaying 256 grey values. Due to the different number of pixels between the camera, (312 x 276), and frame grabber (768 x 512), as well as different aspect ratios (camera formatted to 4:3, frame grabber 3:2) it was necessary to determine a scaling factor. With a known camera-monitor scaling factor and radiographic magnification factor, actual focal spot dimensions could be calculated. A simple method based on measuring the image of a known object size was used.
64
Tuchyna et al
Australas. Phys. Eng. Sci. Med. Vol. 27, No 2, 2004
x A CMOS image sensor method
Focal Spot
Projected cross-hairs from Light Beam Diaphragm
Image Sensor Assembly
Reference Axis Axis of symmetry of the slit
Adjustable x-y Bench
Slit Assembly (b)
(a)
Figure 2. a) System set up for electronic image capture. The slit assembly is seen below the light beam assembly located along the reference axis of the beam, with the image sensor placed below, with its plane perpendicular to the same axis, with the use of the adjustable x-y table. b) Alignment of the slit diaphragm.
A small circular hole of 0.6 mm diameter was made with a high speed drill in the Kodak phosphor. The phosphor was coupled to the photodiode array using silicone grease and exposed at 75 kV and 200 mA. The phosphor was then exposed several times at various focalcamera distances. Exposure time was limited to the standard video frame length of 40ms at 50Hz. The resultant image was sharpened by use of a high pass filter. Edge detection was performed by application of an edge detection algorithm available in Paint Shop Pro. The resultant image was greatly magnified on the computer screen to enable the counting of individual pixels and the number of pixels from edge to edge averaged over two orthogonal dimensions were counted for each image and hence the scale factor calculated. This yielded, 80 pixels = 0.6 mm, so that a scale factor of 1 pixel = 0.0075 mm was determined. The images were found to be circular to within 3 or 4 pixels, translating to an error of r 5%, with the discrepancy possibly attributed to the edge detection algorithm. Although this is not an extremely accurate method to determine the scale factor, it is very simple and fast to perform.
reference axis passed through the centre of the slit diaphragm, as shown in Figure 2. A spirit level was used to check that the face of the slit diaphragm was normal to the reference axis. The image sensor also required precise positioning, with its plane normal to the reference axis. The specially constructed x-y bench jig was used to position the image sensor assembly in directions parallel and perpendicular to the cathode-anode axis. The digital images obtained were difficult to see on the monitor due to the low exposure time of 40 ms, and hence the low light output of the phosphor. Exposures were made on both the large and small foci at settings of 75 kVp and various mA values, to test the effect of mA. Finally, the focal spot slit images were enhanced to facilitate analysis and observation. Dimensions of the focal spot were determined from the focal spot intensity profile and its corresponding FWHM in orthogonal orientations. The image dimensions being magnified were converted to the effective focal spot size width and length, by dividing by the magnification factor defined above. The AS/NZS 4274 standard defines the focal spot value f as a dimensionless quantity with reference to permissible values of width and length when measured with the slit method. An excerpt from the standard reproduced as Table 1, illustrates the focal spot value f along with its range of focal spot width and length values.
Image sensor measurements The measurement of the x-ray focal spot required that the slit be positioned accurately in the central beam. The centre of the projected cross-hairs provided on the light beam diaphragm was used as one end of the reference axis. The slit assembly rested in a purpose built stainless steel paddle, lined with 1mm of lead and supported on a retort stand. The paddle contained a circular opening for the slit assembly, which allowed for easy rotation between images. The slit was positioned at a distance of 30 cm from the focal spot, with a slit detector distance of 42 cm and a corresponding magnification of M = 1.4, along the reference axis so that the
Image enhancement With the camera-frame grabber configuration used, each photodiode is capable of collecting only a limited number of photons during the x-ray exposure, hence generating a small photocurrent per detector pixel. This limitation is primarily dictated by phosphor efficiency and frame length. Since the acquired images did not have a brightness range covering the full range of the digitiser, image enhancement techniques 65
Tuchyna et al
Australas. Phys. Eng. Sci. Med. Vol. 27, No 2, 2004
Nominal focal spot value
possible shift of several pixels does not appear to be critical in determining the nominal focal spot size at the range measured, the accuracy of this technique is limited and more suited to the measurement of larger focal spot sizes at higher mA settings.
Focal spot dimensions Permissible value (mm)
g 0.1
Width
Length
0.10 – 0.15
0.10 – 0.15
0.2
0.20 – 0.30
0.20 – 0.30
0.4
0.40 – 0.65
0.60 – 0.85
p
p
p
1.0
1.00 – 1.40
1.40 – 2.00
1.2
1.20 – 1.70
1.70 – 2.40
1.4
2.20 – 2.90
2.00 – 2.80
1.6
1.60 – 2.10
2.30 – 3.10
1.8
1.80 – 2.30
2.60 – 3.30
p
p
p
x A CMOS image sensor method
Direct film measurements Focal spot slit images were also obtained using the direct film method with the radiographic film, Fuji RX-U, placed at a distance of 106.5 cm from the slit diaphragm. The slit was positioned at a distance of 30 cm from the focal spot with a slit-film distance of 76.5 cm, giving a magnification of M = 2.55, along the reference axis so that the reference axis passed through the centre of the slit diaphragm, as shown in Figure 2. Exposures were made on both the large and small foci at settings of 75 kVp and various mA values as for the image sensor measurements. The films were processed (Protec M45 processor and Agfa developing chemicals at 34oC). The focal spot width and length films were examined on a light box and measured using a 10x magnifying glass with a built in graticule having a scale with divisions of 0.1 mm which were read to the nearest 0.05 mm. All measurements were scaled by the appropriate magnification factor to determine the actual focal spot size.
Table 1. Permissible values of focal spot dimensions for nominal focal spot values according to AS/NZS 4274:1995.
were required to improve visual discrimination of small variations in sample density17-18. Appearance of the captured images was improved using established image enhancement techniques in succession, namely: frame averaging, histogram expansion and neighbourhood averaging19-21. Frame averaging was performed by acquiring and adding together N image frames. For images obtained at 50 and 100 mA, 16 frames were averaged. For images obtained at 200 and 300 mA, 8 frames were averaged. The added images were divided by N to re-scale the pixel data in order that brightness values did not exceed the maximum grey level value of 255. Frame averaging was accomplished using KHOROS, a specialist image analysis program. The averaging of frames in this way improved the signal to noise ratio. Due to the narrow brightness range of the averaged image, typically less than 7% of the full range of 0-255, the image contrast was poor. Linear expansion of contrast was carried out by assigning the darkest pixel value to black and the highest pixel value to white. To further improve the signal to noise ratio, a neighbourhood averaging operation was performed. This involved the computation of each output pixel as a function of a 5 x 5 Gaussian approximated kernel with V = 0.625 pixels. This kernel was implemented by the image analysis program, KHOROS. Although these processes improved the appearance of the captured images, such transformations are likely to distort raw data, hence the focal spot intensity profile obtained from each frame was used to determine focal spot dimensions. The combined effect of image processing could shift the apparent size of an image structure. A shift of 14 pixels would be required to cause an error of 0.1mm (from the scaling factor of 80 pixels = 0.6 mm). For a typical diagnostic x-ray machine of a nominal focal spot size of g!0.4, the range of focal spot dimensions may vary by more than 0.2 mm without affecting the nominal focal spot size. Although the results obtained demonstrate that the
Film-screen measurements The radiographs for film-screen measurements were obtained by using the same set up as for the direct film method described above; however the Fuji RX-U film was used in conjunction with a cassette and Kodak X-Omatic fine screens. The viewing and measurement method was exactly the same as for the direct film.
Results Focal spot size The dimensions of the focal spot were determined from the focal spot intensity profiles of the captured electronic images, and their corresponding FWHMs. Examples of captured focal spot slit images and the corresponding focal spot intensity profiles are shown in figure 3. Images were obtained at exposure settings of 75 kVp, 40 ms over the available range of mA values and a magnification of 1.40. Table 2 gives the focal spot sizes determined from the captured electronic images. Note that the focal spot value f, referred to in tables 2 and 3 is a dimensionless quantity with reference to permissible values of width and length (see table 1). The analysed focal spot dimensions for the Odel Polaire x-ray machine show a nominal focal spot of g 1.0 at 50 mA, for measured dimensions of 1.37 mm and 1.82 mm, for the width and length respectively. At 100 mA, a nominal focal spot of g 1.0 was also determined, although the dimensions of width and length are somewhat greater at 1.39 mm and 1.92 mm, respectively. At the 200 mA and 300 mA tube current settings a nominal focal spot of g 2.2 is obtained and again it is 66
Tuchyna et al
Australas. Phys. Eng. Sci. Med. Vol. 27, No 2, 2004
measurement methods are shown in Table 3. It was noted that the dimensions are smallest for the direct film method, increasing slightly for the screen-film and image sensorphosphor methods. Due to the range of permissible focal spot size widths and lengths used to define the nominal focal spot size (see Table 1), excellent agreement of the nominal focal spot size was obtained for the slit tool in combination with direct film, film-screen or the CMOS image sensor. The accuracy of the image sensor system appears to be about r10% when the actual width and length dimensions are compared to the direct film method.
b) Length
a) Width
0
1
2
3
4
5mm
Approximate Scale
Brightness level
FSIP 250 200 150 100 50 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
FSIP 250 200 150 100 50 0 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Discussion In conclusion, a method based on the slit tool was developed and tested under laboratory conditions for the measurement of focal spot size. We have demonstrated that a digital system based on an opto-coupled phosphor and VVL 1011C CMOS monochrome image sensor can acquire x-ray images of a radiographic focal spot. The acquired images were however of relatively poor quality and required enhancement to achieve satisfactory visual observation. Enhancement of the captured images substantially improved the image quality and enabled focal spot dimensions to be determined. However, the process of enhancement limits the measurement accuracy of the method by introducing image distortion such as pixel shifting. The greater dimensions obtained for the same machine-selected focal spot at the higher mA values is possibly attributable to the greater photon flux and hence a greater amount of light output from the phosphor at the higher mA setting. Although the dimensions are slightly greater, the nominal focal spot size is unaffected. Several improvements to the described method could be incorporated in a further model. Firstly, the camera appears to be a limiting component in the set-up, mainly due to its synchronous signals resulting in short frame grab times. Either programming of the frame grabber or purchase of an asynchronous camera would eliminate this major problem. Further, a larger photodiode array would permit the use of greater magnifications, allowing the imaging of small focal spot sizes. A reduced pixel size would also improve the system resolution. CMOS image sensors with a larger photodiode array and a reduced pixel size of 7.5 Pm x 7.5 Pm have recently become available. In addition, a reducing taper fibre optic coupler could be used, which would also improve resolution. The choice of phosphor appears to be good, although system resolution could be improved by using a pure Gd2O2S:Tb coating instead of using it in its screen form. For routine use, integration of the trigger circuit to the HV generator, requiring only a BNC type connection, would simplify the model. Finally, the use of a dedicated package capable of image enhancement with minimal image distortion and generation of focal spot intensity profiles would enhance the user interface. About 300 exposures of 0.5s average duration were made over the course of the project. No adverse radiation damage effects were noted.
Distance across image (mm)
Figure 3. Cross sectional images of a focal spot determined using the image sensor showing a) width and b) length as well as the corresponding focal spot intensity profile (FSIP) for exposure settings of 75 kVp, 100 mA and 40 ms. The displayed images were enhanced with frame averaging, histogram expansion and neighbourhood averaging as described in the text. Note, the included scale is approximate only.
Tube current (mA) 50 100 200 300
Focal-spot width (mm) 1.37 1.39 2.65 2.70
x A CMOS image sensor method
Focal-spot length (mm) 1.82 1.92 3.16 3.36
Nominal ƒ 1.0 1.0 2.2 2.2
Table 2. Focal spot sizes determined from images captured by the image sensor. Dimensions for the images were obtained from the FWHM of corresponding focal spot intensity profiles.
noted, that the dimensions at 300 mA are somewhat greater than at 200 mA. The dimensions of the focal spot, and hence the nominal focal spot size, determined from the X-ray images for the direct film and screen-film methods are shown in Table 3. Images were obtained at 75 kV and tube current settings of 100 and 300 mA. Significantly increased exposure times were necessary in order to obtain visible images. Increased exposure times were obtained by repeated exposures. A magnification value of 2.55 was used as opposed to the magnification of 1.40 that was used with the image sensor device. The increased magnification value was obtained for a longer focus-to-film distance, which was not possible with the image sensor due to the small (4.6 mm x 6.1 mm) image sensor area. The measurements and subsequent analysis showed that nominal focal spot sizes of g 1.0 and g 2.2 are obtained for the direct film and screen-film methods. A comparison of the measured focal spot sizes for the three 67
Tuchyna et al
Australas. Phys. Eng. Sci. Med. Vol. 27, No 2, 2004
x A CMOS image sensor method
Tube current (mA)
Focal spot width (mm)
Focal spot length (mm)
Nominal focal spot size g
Method
100 100 100
1.35 1.36 1.39
1.68 1.71 1.92
1.0 1.0 1.0
Direct film Screen film Image sensor
Tube surrent (mA)
Focal spot width (mm)
Focal spot length (mm)
NominalfFocal spot size g
Method
300 300 300
2.60 2.67 2.70
3.23 3.32 3.36
2.2 2.2 2.2
Direct film Screen film Image sensor
a) Small focal spot
b) Large focal spot Table 3. Comparison of focal spot size slit image measurements obtained under direct, screen-film and image sensor conditions for a) the small focal spot at 100 mA and b) the large focal spot at 300 mA. AS/NZS 4274, 1995. 10. Nickoloff, E. L., Donnelly, E., Eve, L. and Atherton, J.V., Mammographic resolution: Influence of focal spot intensity distribution and geometry, Med. Phys., 17(3), 436-447, 1990. 11. Schiabel, H., Ventura, A. and Frere, A. F., A formal study of lateral magnification and its influence on mammographic imaging sharpness, Med. Phys., 21(2), 271-276, 1994. 12. Yaffe, M. J., and Rowlands, J. A., X-ray detectors for digital radiography, Phys. Med. Biol. 42, 1-39, 1997. 13. Kandarakis, I., Cavouras, D., Panayiotakis, G., Agelis, T., Nomicos, C. D., Giakoumakis G. E, X-ray induced luminescence and spatial resolution of La2O2S:Tb phosphor screens, Phys. Med. Biol, 41, 297-307, 1996. 14. Giakoumakis, G.E., Nomicos, C.D., Yiakoumakis, E.N., Katsarioti, M.C., Kalikatsos, J.A., Rovithi, M., Panayiotakis, G. S., Evangelou, E. K., Y2O2S:Eu phosphor screens evaluation, Med. Phys. 20(1), 79-83, 1993. 15. Giakoumakis, G. E., Nomicos, C. D., and Sandilos, P. X., Absolute efficiency of Gd2O2S:Tb screens under fluoroscopy conditions, Phys. Med. Biol, 34, 673-8, 1989. 16. Giakoumakis, G. E. and Nomicos, C. D., Absolute efficiency of Y2O2S:Tb screens under fluoroscopy conditions, J.Appl.Phys. 58(7), 2742-2745, 1985. 17. Russ, J., The image processing handbook, 2nd Ed., CRC Press, Boca Raton, Florida, 1994. 18. Bankman, I. N., Handbook of Medical Imaging, Academic Press, San Diego, California, 2000. 19. de Graff, C. N., and Viergever, M.A., Information processing in medical imaging, Plenum 20. Press, New York, 1988. 21. Frei, W., Image enhancement by image hyperbolization, Comp. Graph. Image Process, 6, 286-294, 1977. 22. Hall, E. H., Almost uniform distributions from image enhancement, IEEE Trans. Comp. C-23(2), 207-208, 1974.
Acknowledgements We would like to thank Dr R I Price and Dr R A Fox for comments on the manuscript.
References 1. Lu, T., Udpa, S. S. and Udpa, L. Tomographic reconstruction using optoelectronic architecture, IEEE Volume 4, 1991. 2. Muirhead, I. T., Developments in CMOS camera technology, IEE. 1994. 3. Lake, D., Why CMOS imagers are the rage: A revolution in the making, Advanced Imaging Magazine, 12-18, 1995. 4. Fossum, E. R., CMOS image sensors: Electronic camera on a chip, IEEE 1995. 5. Denyer, P. B., Renshaw, D., Wang Guoyu and Lu Mingying, A single chip sensor and image processor for fingerprint verification, Proceedings of IEEE Custom Integrated Circuits Conference, San Diego, California, May 9-12, 1993. 6. Speller, R. D., Martinez-Davalos, A., and Farquharson, M., A CCD based focal spot camera, Phys. Med. Biol. 40, 315-321, 1995. 7. Rong, X. J., Krugh, K. T., Shepard, S. J. and Geiser, W. R. Measurement of focal spot size with slit camera using computed radiography and flat-panel based digital detectors, Med. Phys. 30(7): 1768 – 75, 2003 8. Everson, J. D. and Gray, J. E., Focal spot measurement: Comparison of slit, pinhole and star resolution pattern techniques, Radiology, 165, 261-264, 1987. 9. Australian/New Zealand Standard (AS/NZS), X-ray tube assemblies for medical diagnosis-Characteristics of focal spots,
68