Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Edge detection wikipedia , lookup
BSAVE (bitmap format) wikipedia , lookup
Computer vision wikipedia , lookup
Indexed color wikipedia , lookup
Charge-coupled device wikipedia , lookup
Hold-And-Modify wikipedia , lookup
Stereoscopy wikipedia , lookup
Image editing wikipedia , lookup
Medical imaging wikipedia , lookup
Stereo display wikipedia , lookup
Digital Radiography – Chapter 11 Adjuncts to Radiology – Chapter 12 Brent K. Stewart, PhD, DABMP Lois Rutz, M.S. Radiation Safety Engineering, Inc. a copy of Brent Stewart’s unmodified lecture may be found at: http://courses.washington.edu/radxphys/PhysicsCourse04-05.html Take Away: Five Things You should be able to Explain after the DR/Adjuncts Lecture The various types of detectors used in digital imaging (e.g., scintillators, photoconductors, etc.) The differences between the various technologies used for digital radiography (e.g., CR, indirect and direct DR) Benefits of each type (e.g., resolution, dose efficiency) Why digital image correction and processing are necessary or useful and how they are executed The various types of adjuncts to radiology (e.g., DSA or dual-energy imaging), what issue they are trying to resolve, mechanism exploited and end result Why Digital/Computed Radiography Limitations on Film/Screen radiography Screen/Film system is image receptor and display Image characteristics depend on Screen/Film and Film processing. Modification of image difficult to control (e.g. development temperature). Image appearance depends on technique settings. Image quality cannot be repaired after development. Retake only solution to poor I.Q. Why Digital/Computed Radiography cont. Screen/film dynamic range 2 to 2.5 orders of magnitude. Different applications require different screen/film combinations. Only one “original” image. Films often “go missing” from ER or ICU and never are archived. Copies expensive, have inconsistent quality, and often are nondiagnostic. Archive space expensive, often remote. Digitizing film is only way to move images to PACS. How does Digital/Computed Radiography solve these problems? • Decouples imaging chain components. • Detector, image processing, display all “independent” entities. • Independent in design but not in application. • Detector can make use of extended dynamic range. • Solid state detectors have improved DQE. • Electronics can apply corrections to input signals. • In particular, over/under exposure can be corrected, reducing retakes. How does Digital/Computed Radiography solve these problems? Cont. • Image processing can modify and enhance raw (preprocessed) data. • Images can be displayed on workstations which permit interactive display processing. • Image data is stored digitally. “Original image” is available everywhere and at any time. CR vs. DR CR also known as a Photostimulable Phosphor system. CR uses an imaging plate similar to an intensifying screen as the receptor. CR systems are indirect digital systems. Indirect systems convert x-radiation to the final digital image through one or more stages. DR digital radiography Uses a fixed detector such as amorphous selenium plate as the receptor. Can be a direct or an indirect digital system. When direct it is sometimes called DDR for direct digital radiography CR Detector or Imaging Plate (IP) is essentially a type of intensifying screen. IP can be used in any bucky or table-top system. IP is relatively robust. Requires same care as intensifying screens. Process is indirect. X-ray creates excitation center. Plate reader uses red light to stimulate centers to release blue light. Blue light is directed to a photo-electric transducer (pmt or other). Electric signal digitized to make raw image. CR and DR Systems Image Production in CR/DR Systems Radiation through the patient creates a latent image on the receptor. Receptor is “read” by some process and latent image is converted to an electronic signal. Signal is processed. Signal (analog) is converted via ADC to a bit value in a digital matrix. Digital image is processed. Processing is related to acquisition system characteristics. Processing is related to desired image information. Digital matrix is displayed on a video screen or printed to paper or film. Signal Processing Primarily to accommodate variations in the detector/electronics components. Involves corrections for dead space, non-uniformities, defects. Could be developed to compensate for MTF losses. All systems, PSP or Direct, do some sort of processing and scaling. Ultimate goal is to present the image processing module with “true” image pixels. Digital Image Correction Interpolation to fill in dead pixel and row/column defects Subtracting out average dark noise image Davg(t)(x,y) Differences in detector element digital values for flat field Make corrections for each detector element (map) Gain image: G(x,y) =G’(x,y) - Davg(t)(x,y); Gavg =(1/N) ∙ G(x,y) I(x,y) = Gavg ∙ [Iraw(x,y) - Davg(t)(x,y)] / G(x,y) Done for DR and in a similar manner for CT (later) Not performed for CR on a pixel by pixel basis, although there are corrections on a column basis for differences in light conduction efficiency in the light guide to the PMT Digital Image Correction c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2 nd ed., p. 310. Detectors In order to understand signal processing we need to learn about the detectors. Photo Stimulable Phosphor Plates Photoconductive materials. Detector consists of a receptor material (e.g. BaF(H)Eu), and a set of signal readout and conversion electronics. Receptor responsible for the DQE. Rest of the system contributes to noise, resolution,dynamic range. Detectors in Digital Imaging (1) Gas and solid-state detectors Energy deposited to e- through Compton and photoelectric interactions Gas detectors – apply high voltage across a chamber and measuring the flow of eproduced by ionization in the gas (typically high Z gases like Xenon: Z=54, K-edge = 35 keV) Were used in older CT units c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p.32. Detectors in Digital Imaging (2) Solid-state materials Electrons arranged in bands with conduction band usually empty Solid-state detectors Scintillators – some deposited energy converted to visible light Photoconductors – charge collected and measured directly Photostimulable phosphors – energy stored in electron traps c.f. Yaffe MJ and Rowlands JA. Phys. Med. Biol. 42 (1997), p. Elements of Digital Radiology, p. 10. Detectors in Digital Imaging (3) c.f. Yaffe MJ and Rowlands JA. Phys. Med. Biol. 42 (1997), p. Elements of Digital Radiology, p. 9. Computed Radiography (CR) Photostimulable phosphor (PSP) Barium fluorohalide: 85% BaFBr:Eu + 15% BaFI:Eu e- from Eu2+ liberated through absorption of x-rays by PSP Liberated e- fall from the conduction band into ‘trapping sites’ near F-centers By low energy laser light (700 nm) stimulation the e- are re-promoted into the conduction band where some recombine with the Eu3+ ions and emit a blue-green (400500 nm) visible light (VL) c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 295. Computed Radiography (CR) System (1) Imaging plate (IP) made of PSP is exposed identically to SF radiography in Bucky IP in CR cassette taken to CR reader where the IP is separated from cassette IP is transferred across a stage with stepping motors and scanned by a laser beam (~700 nm) swept across the IP by a rotating polygonal mirror Light emitted from the IP is collected by a fiber-optic bundle and funneled into a photomultiplier tube (PMT) PMT converts VL into e- current c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 294. Computed Radiography (CR) System (2) Electronic signal output from PMT input to an ADC Digital output from ADC stored Raster swept out by rotating polygonal mirror and stage stepping motors produces I(t) into PMT which eventually translates into the stored DV(x,y): PMT→ADC→RAM IP exposed to bright light to erase any remaining trapped e- (~50%) IP mechanically reinserted into cassette ready for use 200mm and 100mm pixel size (14”x17”: 1780x2160 and 3560x4320, respectively) c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 294. Indirect Flat Panel Detectors Use an intensifying screen (CsI) to generate VL photons from an x-ray exposure Light photons absorbed by individual array photodetectors Each element of the array (pixel) consists of transistor (readout) electronics and a photodetector area The manufacture of these arrays is similar to that used in laptop screens: thin-film transistors (TFT) c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 301. Charged-Coupled Devices (CCD) Form images from visible light Videocams & digital cameras Each picture element (pixel) a photosensitive ‘bucket’ After exposure, the elements electronically readout via ‘shiftand-read’ logic and digitized Light focused using lenses or fiber-optics Fluoroscopy (II) Digital cineradiography (II) Digital biopsy system (phosphor screen) 1K and 2K CCDs used c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., pp. 298-299. Direct Flat Panel Detectors Use a layer of photoconductive material (e.g., α-Se) atop a TFT array e- released in the detector layer from x-ray interactions used to form the image directly X-ray→e-→TFT → ADC→RAM High degree of e- directionality through application of E field Photoconductive material can be made thick w/o degradation of spatial resolution Photoconductive materials Selenium (Z=34) CdTe, HgI2 and PbI2 Indirect Flat Panel Detector (for comparison) Direct Flat Panel Detector c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 304. Thin-Film Transistors (TFT) After the exposure is complete and the e- have been stored in the photodetection area (capacitor), rows in the TFT are scanned, activating the transistor gates Transistor source (connected to photodetector capacitors is shunted through the drain to associated charge amplifiers Amplified signal from each pixel then digitized and stored X-ray→VL→e-→ADC→RAM c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 301. Resolution and Fill Factor Dimension of detector element largely determines spatial resolution 200mm and 100mm pixel size typical For dimension of ‘a’ mm - Nyquist frequency: FN = 1/2a If a = 100mm → FN = 5 cycle/mm Fill factor = (light sensitive area)/(detector element area) Trade-off between spatial resolution and contrast resolution c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2 nd ed., p. 303. Image Digitization and Processing After acquisition and correction of raw data, the image is ready for display processing. The image data consists of a matrix of numbers. Each pixel is one matrix point. Each gray scale is a digital value. For example: a matrix can have 1024 x 1024 pixels and each pixel will have a value from 0 to 1024. Each value is related to the radiation exposure which created that pixel. Digital Storage of Images Usually stored as a 2D array (matrix) of data, I(x,y): I(1,1), I(2,1), … I(n,m-1), I(n,m) Each minute region of the image is called a pixel (picture element) represented by one value (e.g., digital value, gray level or Hounsfield unit) Typical matrices: CT: 512x512x12 bits/pixel CR: 1760x2140x10 bits/pixel DR: 2048x2560x16 bits/pixel c.f. Huang, HK. Elements of Digital Radiology, p. 8. Image Processing Image data is scaled to present image with appropriate gray scale (O.D.) values regardless of the actual radiation used to produce the image. Image data is frequency enhanced around structures of importance. Process involves mathematical filters. Image data is display processed to give desired contrast and density. Process involves re-mapping along a chosen display (“H&D”) curve Generic Display Processing Different manufacturers may use different versions of generic image processing methods. E.g. Musica, Ptone All describe means of scaling and modifying image appearance. Different manufacturers use different exposure indicators. E.g. EI, S, IgM All describe the relationship between the exposure to the detector and the pixel value. Generic Elements of Display Processing Exposure Recognition. Signal Equalization: Adjust regions of low/high signal value Grayscale Rendition Adjust for high/low average exposure Convert signal values to display values Edge Enhancement: Sharpen edges M. Flynn, RSNA 1999 Image Processing normalized histogram comparison fraction of pixels in ROI 0.025 0.02 0.015 ph1.2 ptx case 4 0.01 0.005 0 0 0.2 0.4 0.6 normalized pixel value 0.8 1 Computed Radiography (CR) System (3) IP dynamic range = 104, about 100x that of S-F (102) Very wide latitude → flat contrast Image processing required: Enhance contrast Spatial-frequency filtering CR’s wide latitude and image processing capabilities produce reasonable OD or DV for either under or overexposed exams Helps in portable radiography: where the tight exposure limits of S-F are hard to achieve Underexposed → ↑ quantum mottle and overexposed → unnecessary patient dose c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 296. Unsharpmasked Spatial Frequency Processing c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2 nd ed., p. 313. Global Processing Most common global image processing: window/level Global processing algorithm I’(x,y) = c ∙ [I(x,y) – a]: essentially y = mx + b Level (brightness) set by a Window (contrast) set by c I’ = [2N/ww]∙[I-{wl-(ww/2)}], where ww = window width and wl = window level Need threshold limits when max/min [2N-1, 0] digital values encountered If I’(x,y) > Tmax→I’(x,y) = Tmax If I’(x,y) < Tmin→I’(x,y) = Tmin c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., pp. 92 and 311. Image Processing Based on Convolution Convolution: Ch. 10 - Image Quality and Ch. 13 - CT Defined mathematically as passing a N-dimensional convolution kernel over an Ndimensional numeric array (e.g., 2D image or CT transmission profile) At each location (x, y, z, t, ...) in the number array multiply the convolution kernel values by the associated values in the numeric array and sum Place the sum into a new numeric array at the same location c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 312. Image Processing Based on Convolution Delta function kernel 0 0 0 0 1 0 0 0 0 Blurring kernel (normalization) also known as low-pass filter 1/9 1/9 1/9 1/9 1/9 1/9 1/9 1/9 1/9 Edge sharpening kernel -1 -1 -1 -1 9 -1 -1 -1 -1 c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 313. Image Processing Based on Convolution Convolution kernels can be much larger than 3 x 3, but usually N x M with N and M odd Can also perform edge sharpening by subtracting blurred image from original → high-frequency detail (harmonization) The edge sharpened image can then be added back to the original image to make up for some blurring in the original image: CR unsharpmasking - freq. processing The effects of convolution cannot in general be undone by a ‘de-convolution’ process due to the presence of noise, but a deconvolution kernel can be applied to produce an approximation: 19F MRI Median and Sigma Filtering Convolution of an image with a kernel where all the values are the same, e.g. (1/NxM), essentially performs an average over the kernel footprint Smoothing or noise reduction This can make the resulting output value susceptible to outliers (high or low) Median filter: rank order values in kernel footprint and take the median (middle) value Sigma filter: set sigma (s) value (e.g., 1) and throw out all values in kernel footprint > m + s or < m – s and then take the average and place in output image Multiresolution/Multiscale Processing and Adaptive Histogram Equalization (AHE) Some CR systems (Agfa/Fuji) make use of multiresolution image processing (AKA unsharpmasking) to enhance spatial resolution Wavelet or pyramidal processing on multiple frequency scales Histogram equalization re-distributes image digital values to uniformly span the entire digital value range [2N-1,0] to maximize contrast AHE does this on a spatial sub-region basis in an image rather than the entire image Fuji ‘Dynamic Range Control’ (DRC) a version of AHE that operates on sub-regions of digital values Histogram Equalization Properly Exposed Image Under-exposed Image Over-exposed Image Histogram Equalized Image c.f. http://www.wavemetrics.com/products/igorpro/imageprocessing/imagetransforms/histmodification.htm Global and Adaptive Histogram Equalization The following images illustrate the differences between global and adaptive histogram equalization. MR image with the corresponding gray-scale histogram. The histogram has a peak at minimum intensity consistent with the relatively dark nature of the image. Global histogram equalization and the final gray-scale histogram. Comparing the results with the figure above we can see that the distribution was shifted towards higher values while the peak at minimum intensity remains. Adaptive histogram equalization shows better contrast over different parts of the image. The corresponding gray-scale histogram lacks the mid-levels present in the global histogram equalization as a result of setting a high contrast level. c.f. http://www.wavemetrics.com/products/igorpro/imageprocessing/imagetransforms/histmodification.htm Contrast vs. Spatial Resolution in Digital Imaging S-F mammography can produce images w/ > 20 lp/mm According to Nyquist criterion would require 25 mm/pixel resulting in a 7,200 x 9,600 image (132 Mbytes/image) Digital systems have inferior spatial resolution However, due to wide dynamic range of digital detectors and image processing capabilities, digital systems have superior contrast resolution c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 315. Digital Imaging Systems and DQE Remember the equation for 2 DQE(f)? k MTF (f ) DQE(f) = N NPS(f ) How can we account for this? Both CR and the screens in film/screens made thin Film higher spatial resolution than CR DQE higher for α-Si systems using CsI and Gd2O2S rather than α-Se (mean x-ray E & Z) α-Si DQE falling off more rapidly than α-Se (geometry) α-Si DR α-Se DR Digital versus Analog Processes & Implementation Although some of the previous image reception systems were labeled ‘digital’, the initial stage of those devices produce an analog signal that is later digitized CR: x-rays→VL→PMT→current→voltage→ADC CCD, direct & indirect digital detectors: stored e- → ADC Benefits of CR Same exam process and equipment as screen-film radiography Many exam rooms serviced by one reader Lower initial cost Benefits of DR Throughput ↑: radiographs available immediately for QC & read Patient Dose Considerations Over and underexposed digital receptors produce images with reasonable OD or gray scale values As overexposure can occur, need monitoring program CR IP acts like a 200 speed S-F system wrt. QDE Use the CR sensitivity (‘S’) number to track dose Bone, spine and extremities: 200 Chest: 300 General imaging including abdomen and pelvis: 300/400 Flat panel detectors can reduce radiation dose by 2-3x as compared with CR for the same image quality due to ↑ quantum absorption efficiency & conversion efficiency Using the CR Sensitivity Number to Track Dose S-number Dashboard Main Exams 500 450 Sep-03 Baseline Mar-04 Apr-04 400 S-number 350 300 Target Values 250 Fixed Chest - [255-345] 200 Bone, Spine & Ext [170-230] 150 100 General Imaging incl. Abdomen - [340-460] 50 0 MFEM MC2 MC5 MCH2 Exam Code MKUB MPELV Huda Ch6: Digital X-ray Imaging Question 12. Photostimulable phosphor systems do NOT include: A. Analog-to-digital converters B. Barium fluorohalide C. Light detectors (blue) D. Red light lasers E. Video cameras Huda Ch6: Digital X-ray Imaging Question 11. Which of the following x-ray detector materials emits visible light: A. Xenon B. Mercuric iodide C. Lead iodide D. Selenium E. Cesium iodide Raphex 2002 Question: Digital Radiography D47. Concerning computed radiography (CR), which of the following is true? A. Numerous, small solid-state detectors are used to capture the x-ray exposure patterns. B. It has better spatial resolution than film. C. It is ideal for portable x-ray examinations, when phototiming cannot be used. D. It is associated with high reject/repeat rates. E. The image capture, storage, and display are performed by the receiver. Huda Ch6: Digital X-ray Imaging Question 13. Photoconductors convert x-ray energy directly into: A. Light B. Current C. Heat D. Charge E. RF energy Huda Ch6: Digital X-ray Imaging Question 15. Which of the following does NOT involve image processing: A. Background subtraction B. Energy subtraction C. Histogram equalization D. K-edge filtering E. Low-pass filtering Huda Ch6: Digital X-ray Imaging Question 14. Processing a digital x-ray image by unsharpmask enhancement would increase the: A. Bit depth per pixel B. Matrix size C. Patient dose D. Visibility of edges E. Limiting spatial resolution Adjuncts and other interesting stuff Geometric (Linear) Tomography c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 318. With the advent of CT, geometric tomography has only limited clinical utility where only one or a few planes of objects with high contrast are desired, e.g., IVP Desired slice through patient set at pivot point (focal plane) The tomographic process blurs out regions outside the focal plane, but still contributes to overall loss of contrast Larger tomographic angles result in a lessening of out of plane contributions High dose, comparable to CT for many tomographic slices Digital Tomosynthesis Improved version of geometric tomography where a digital detector saves an image at each of several tube angles This allows reconstruction of multiple planes through the object through shifting the various images through a certain distance before summing them Much more dose efficient, but still suffers from out of plane blurring effects Either CR or DR used c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 320. Temporal Subtraction Digital Subtraction Angiography (DSA) – usually 1K resolution Mask (background) subtracted from images during/post contrast injection: Δ < 1% trans. visualized Motion can cause misregistration artifacts Digital value proportional to contrast concentration and vessel thickness Is = ln(Im) – ln(Ic) = mvessel ∙ tvessel Temporal subtraction works best when time differences between images is short Possible to spatially warp images taken over a longer period of time c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 322. Dual-Energy Subtraction Exploits differences between the Z of bone (Zeff ≈ 13) and soft tissue (Zeff ≈ 7.6) Images taken either at two different kVp (two-shot) One image (one-shot) taken with energy separation provided by a filter (sandwich) Iout = loge(Ilow) – R ∙ loge(Ihigh), where R is altered to produce soft-tissue predominant or bone predominant images GE Chest DR @ SCCA c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2nd ed., p. 324. Dual-Energy Subtraction c.f. Bushberg, et al. The Essential Physics of Medical Imaging, 2 nd ed., p. 325. Huda Ch6: Digital X-ray Imaging Question 22. The matrix size in a DSA image is typically: A. 128 x 128 B. 256 x 256 C. 512 x 512 D. 1024 x 1024 E. 2048 x 2048 Huda Ch6: Digital X-ray Imaging Question 25. Changing the DSA matrix from 10242 to 20482 would NOT increase the: A. Data digitization rate B. Data storage requirement C. Image processing time D. Spatial resolution E. Pixel size Raphex 2003 Question: Digital Radiography D51. A flat panel digital radiographic detector has a square 20 x 20 cm image receptor field. The full field of the detector is coupled to a nominal 2048 x 2048 CCD array. The relative spatial resolution (lp/mm) when going from a 20 x 20 cm to a 10 x 10 cm field of view is: A. Four times better B. Twice as good C. The same D. Half as good E. One fourth as good Huda Ch6: Digital X-ray Imaging Question 17. The Nyquist frequency for a 1K digital photospot image (25 cm image intensifier diameter) is: A. 1 lp/mm B. 2 lp/mm C. 4 lp/mm D. 8 lp/mm E. 10 lp/mm FN (lp/mm) = 1/2a = 1/2(1024 lines/250 mm) = 2.048 ≈ 2 Digital Representation of Data (1) Bits, Bytes and Words Smallest unit of storage capacity = 1 bit (binary digit: 1 or 0) Bits grouped into bytes: 8 bits = byte Word = 16, 32 or 64 bits, depending on the computer system addressing architecture Computer storage capacity is measured in: kilobytes (kB) - 210 bytes = 1024 bytes a thousand bytes megabytes (MB) - 220 bytes = 1024 kilobytes a million bytes gigabytes (GB) - 230 bytes = 1024 megabytes a billion bytes terabytes (TB) - 240 bytes = 1024 gigabytes a trillion bytes Digital Representation of Data (2) Digital Representation of Different Types of Data Alphanumeric text, integers, and non-integer data Storage of Positive Integers In general, n bits have 2n possible permutations and can represent integers from 0 to 2n-1 (the range usually denoted with square brackets): n bits represents 2n values with range [0, 2n-1] 8 bits represents 28 = 256 values with range [0, 255] 10 bits represents 210 = 1024 values with range [0, 1023] 12 bits represents 212 = 4096 values with range [0, 4095] 16 bits represents 216 = 65,536 values with range [0, 65535] Conversion of Analog Data to Digital Form The electronic measuring devices of medical scanners (e.g., transducers and detectors) produce analog signals Analog to digital conversion (analog to digital converter – ADC) ADCs characterized by sampling rate or frequency (e.g., samples/sec – 1 MHz) number of bits output per sample (e.g., 12 bits/sample = 12-bit ADC) c. f. Bushberg, et al., The Essential Physics of Medical Imaging, 2nd ed., p. 69. Periodic Table of the Elements c.f. http://www.ktf-split.hr/periodni/en/