NOTES Teaching Digital Processing and Display Quinn B Carroll, MEd, RT WCEC Conference, 2015

Size: px

Start display at page:

Download "NOTES Teaching Digital Processing and Display Quinn B Carroll, MEd, RT WCEC Conference, 2015"

Derek Jones
5 years ago
Views:

1 NOTES Teaching Digital Processing and Display Quinn B Carroll, MEd, RT WCEC Conference, Teaching Pitfalls: Accuracy Vs. Simplicity 2. Teaching Decisions: 1. Is it relevant? (Not: Is it in my comfort zone?) 2. What is the empirical, experimental truth? (Not: What is the majority opinion, even among authors) 3. Is it more important to be accurate, or to be comprehensible ( keep it simple )? 4. What terminology, analogies, illustrations, etc. will best convey this concept? 3. Lab Pitfalls in Demonstrating Contrast Δ: a. NEVER use background density b. Use two relatively homogeneous density areas within the anatomy c. Re calculate 15% for each step increase in kvp 4. Using Ratios for Contrast Densities 3 and 1: Add fog at +1 Example: Effect of fog on subject contrast: 3/1 and 4/2 Vs. 3 1 and 4 2 Examples: Effect of SID or mas on subject contrast: 2/1 and 4/2 Vs. 2 1 and For subject contrast within the remnant beam reaching the IR, it s the proportions of interactions that matters, not the raw numbers of interactions 6. Using Ratios Vs. Percentages Percentage Change Ratio Change +100% 2X 50% ½ +50% 3/2 (1 ½ X) 33% 2/3 +33% 4/3 (1 1/3 X) 25% ¾ +25% 5/4 (1 ¼ X) 20% 4/5 The New Deviation Index 7. The deviation index changes by +1.0 for each +25% (increase in exposure), and by 1.0 for each 20% change. These are proportional ratios: +25% = 5/4 20% = 4/5

2 8. Deviation Exposure Recommended Index Deviation Descriptor Action. >+3.0 >100% high Excessive Pt No Repeat unless Exposure saturation occurs. Mgt follow up. +1 to % high Overexposure Repeat only if saturation occurs 0.5 to % to +25% Target Range 1 to % target Underexposed Consult radiologist for repeat < 3.0 < 50% target Excessive Repeat underexposure (Mottle is certain) 9. Exposure less than 80% of the EI T should not be repeated unless a radiologist finds the level of mottle in the image unacceptable. Only exposures less than 40% of the EI T should be assumed to be repeatable. No overexposed digital image should be repeated, no matter the EI, unless saturation has occurred! 10. Saturation: Takes at least eight to ten times normal exposure. Corresponds to an EI 3000 for the Kodak system, or an S number of 25 for Fuji. Results in flat black appearance of overexposed portion of image. This is not fog, but a loss of data, with absolutely no details present 11. Inappropriate Use of DI: Even if images being produced clinically have corresponding DI s well within the target range, the clinical techniques used may still not be appropriate. One can just as readily achieve an acceptable DI for an AP L spine view with 65 kvp as with 85 kvp; evidence of underpenetration and concomitant excess patient exposure with lower kvp may be windowed and leveled out in a digital image. Similarly, poor collimation, unusual patient body habitus, the presence of prosthetic devices, or the presence of gonadal shielding in the image may raise or lower DI s (depending on the exam and projection) and perhaps hide an inappropriate technique. AAPM Task Group 16, JMP, Vol. 36, No. 7, July Why We Need Noise 13. Medical physicists have been emphasizing the importance of Signal to Noise ratio for years More important now, with digital imaging We still do not have either term in our standard definitions 2011 draft of ASRT curriculum guide mentions noise twice: 1) Under Adequate Exposure Level, implying that it is limited only to quantum mottle, and 2) Under Spatial Resolution 15. If one had to choose a single current category to place mottle under, contrast or gray scale would be most fitting, as demonstrated here Physicists use low contrast dots to measure the level of mottle present 1. How should we define noise? Anything which obstructs visibility of image details c/o Dr. Anthony Wolbarst The Physics of Radiology

3 2. Image Noise: 16 types! 3. Quotes from Dr. Anthony Wolbarst s The Physics of Radiology : From a broader perspective, noise is anything in an image that detracts from its clinical usefulness Imaging detector devices generate weak electric signals and the number of electrons actually involved at any instant will vary about the average according to Poisson statistics. Sometimes more noticeable is the noise injected into the sensitive detector electronics by external sources, such as lightening storms, sparking of machinery, and some fluorescent lamps Imaging devices can contribute to noise in a variety of ways. Imperfections in computer reconstruction algorithms may lead to abnormalities in images produced 4. In this scheme, where do we place extraneous artifacts, grid lines, quantum mottle, or electronic noise?

4 5. Proposed Categories: All negative visibility factors in 1 column 6. Proposed general categories: 7. Three Domains for Digital Processing: 1. In the spatial domain, pixels are sorted by their location 2. In the intensity domain, pixels are sorted by their pixel value (brightness, density) 3. In the frequency domain, objects are sorted by their size Manufacturer chooses by comparing results for a particular objective 8. Three general ways of sorting a digital image:

5 9. Default Processing: Preprocessing: 1. Field Uniformity Corrections 2. Noise and Del Drop out Corrections 3. Image and Histogram Analysis 4. Rescaling (Normalization) Postprocessing: 5. Gradation Processing (LUTs) 6. Detail Processing 7. Preparation for Display IMAGE DISPLAYED 8. Operator Adjustments 9. Application of Special Features 10. Rescaling Vs. Gradation Processing: 11. Both use intensity transformation formulas to re map input pixel values as gray levels. Objective of Rescaling is to normalize appearance. Purpose of Gradation (Gradient) Processing is to refine gray scale according to anatomy 12. Rescaling and Gradient Processing both use LUTs What exactly is an LUT? 13. Look Up Tables really are tables, (not graphs). 2 simple columns: Input and Output. Graphs representing LUTs are for human benefit in understanding the formulas (functions) that generate them. The computer deals in numbers, not in graphs. 14. Permanent: Output already calculated once and permanently stored. Variable: Output generated by formula for each input set (exposure) 15. Permanent look up table of Q values is stored in computer. Here: Qmin is always output at 511, Q2 as 512, Q3 as 513 Qmax as For the acquired image histogram, files of data designated as Smin, S2, S3, S4 and so on up to S1024 and finally Smax 17. Incoming pixel values can be made to fit a pre set range of S values, depending on how extremely they are rounded up or down by digitization 18. KEY: The range of S values in the acquired image matches the range of Q values in the master look up table 19. Algorithm remapping S values to Q values: Set Smin = Qmin, S2 = Q2, S3=Q3, Smax = Qmax 20. Since the # of Q values is fixed, the resulting brightness range is always constant 21. In effect, the histogram of the new exposure is re aligned to the stored ideal histogram of the Q range (A to B) 22. Digitization can re map the high, low and average points in the data matching the S range to the Q range: This roughly standardizes Contrast and Gray Scale 23. Rescaling partly corrects for gray scale What the computer cannot do is (C) alter which pixels contain particular values (change pixel counts) This would perfectly match the shape of the acquired histogram to that of the stored ideal histogram 24. Purpose of Gradation Processing: Fine tune or customize contrast and gray scale according to the anatomical procedure All gradation processing takes place in the intensity domain

6 25. In gradation processing we use variable LUTs that are anatomy specific (anatomical LUTs). The rescaled S values, (always the same number set, is fed into the anatomical LUT 26. Those darn physicists: Q Values: Recommend: 1. Q values ( for processing before rescaling) S values 2. QK values ( for processing after rescaling) Q values 3. QP values ( for presentation ) QP values 27. At the control console, selecting the procedure algorithm assigns the anatomical LUT to be used for gradation processing 28. Look up Table A produces medium contrast, Table B produces high contrast from the same data. Note that in the output column, average gray level or center (GC) is the same for both 29. Anatomical LUTs are generated from gradient curve formulas (fxs). LUT function curve for enhancing image contrast: Original Contrast = 12 / 10 = 1.2 Converted Output Contrast = 18 / 7 = The actual look up table is simply a listing of the input values and then the output values resulting from these mathematical operations 31. Gradient curve formulas are usually Intensity Transformation Formulas: Gamma Transformation: s = cr γ Log Transformation: s = clog(1 + r) Image Negative: s = L 1 r In each formula, r represents the input value and s represents the output value 32. The same formulas are applied whenever the radiographer windows the image after initial display: 1. Parameters in formula are changed 2. Formula generates new LUT 3. LUT is read out. Gradient processing used twice: Default processing, windowing 33. γ = window width control, r = Input value, s = Output value, (C = constant affecting the type of change made) 34. As γ decreases, gray scale lengthens, As γ increases, gray scale shortens 35. The type (shape) of curve used by a manufacturer represents the algebraic function (formula) applied to set up the anatomical LUT. Different parameters entered into the formulas change the position or slope of the curve 36. Fuji s Gradient Parameters: GA = amount, GT = type, GC = center, GS = shift of the function curve 37. Consistent Use of Terms: Brightness and Contrast Controls Vs. Window Level and Width: Strictly speaking, High Window Level means a darker image, Window level = opposite of brightness. High Window Width means more gray scale, Window width = opposite of contrast 38. Dynamic Range Compression and Equalization: 39. Dynamic Range Compression (DRC): Two Purposes: 1. Save computer storage space (when Δ not visible) 2. Apply tissue equalization to image (when Δ visible) 40. Bit Depth, Dynamic Range, and Gray Scale: Difference may be thought of as: Bit Depth describing capacity of hardware, Dynamic Range describing settings in the system with its particular installed software, Gray Scale describing the displayed image

7 41. Use of the whole bit depth of the computer increases processing time unnecessarily, so a smaller range is selected from which to build up the image 42. Dynamic Range Control (DRC) Vs. Tissue Equalization Vs. Contrast Equalization Vs. Normalization 43. Limit of Human Eye DR = 32 (2 5 ) 44. To allow windowing, we must be able to double or cut in half both the brightness and contrast of the image several times without running out of dynamic range (or data clipping will occur). Complex features such as subtraction require still more, Typical: 2 10 for CR & DR, 2 12 for MRI, CT 45. Thus, postprocessing capabilities needed for the image set a lower limit on dynamic range 46. Tissue Equalization is DRC operating just below Mathematically, DRC finds the mid point of the gray scale curve, then progressively reduces pixel values above this point, and progressively increases pixel values below it. Applied to a degree visibly affecting the image, this results in truncation (cutting off) of the darkest and lightest densities in the gray scale of the image 48. DRC is better described as Gray Scale Truncation A clipping off of the extreme ends of the dynamic range. Simulates soft tissue technique used with S/F 49. Using DRC for Tissue Equalization: Gray scale truncation: Elimination of darkest and lightest densities in gray scale range results in a grayer looking image! Unlike conventional shortening of GS which results in increased contrast 50. Over applying TE results in loss of details 51. Digital Features: Application and Over application 52. Digital Processing Features: Application and Over application: Windowing Brightness Gray Scale Look Adjustment Edge Enhancement Smoothing Local Area Rescaling: Ex: Underpenetrated [Correction for underexposure] Special Features: Subtraction Tomographic Artifact Suppression Image Stitching 53. Manufacturers use various terms for edge enhancement and smoothing functions. Example: DR unit by GE has a feature called the look of the image, that can be set to normal, hard, or soft. Hard = Edge enhancement, Soft = Smoothing 54. Over use of Smoothing or Edge Enhancement: Over application of either can lead to loss of detail. If an image already has long gray scale or low contrast, applying smoothing in any degree can lead to loss of visibility for fine details (from low local contrast). If an image already possesses high contrast, applying edge enhancement can cause excessive noise 55. Over Use of Edge Enhancement: It is very tempting to overuse the edge enhancement feature or even form a habit of applying it to all images that come up for review. Resulting increase in

8 image noise can reduce exposure latitude such that only a 30% reduction in exposure can manifest unacceptable levels of mottle. Can also cause halo artifact 56. A veterinary knee image shows the halo effect in A, left, (green arrows). The effect was eliminated by disengaging the edge enhancement feature, (B) 57. Applying EE to an image that already possesses adequate contrast can actually obscure critical details and lead to misdiagnosis. Default setting can be too much 58. The degree of edge enhancement or smoothing applied can be customized by the QC technologist (w/ radiologist) 59. Smoothing / Noise Reduction: Useful primarily for electronic noise. May improve moderate amounts of mottle, but CANNOT correct for serious mottle from underexposure 60. Modern units can detect the area of the image where average pixel values are too light, then selectively darken only that portion of the image 61. Rescaling Identified Area: Underpenetrated Function (GE): Area can be identified by build up pattern of consistently low pixel values. All values within this area are then multiplied by a constant [determined by procedure] to darken [or lighten] entire area by a pre determined ratio 62. For the lateral C spine, the original problem is caused by underexposure of the x ray beam in the C7 T2 area. The correction itself is only an adjustment in brightness 63. Thus, the term underpenetrated used by GE is yet another misnomer and very misleading If there were not some penetration of the x ray beam through this area, there would be no data in this area to recover 64. It can also confuse the student into thinking that the computer can somehow restore penetration to the image a completely false notion 65. GE s underpenetrated fx is better described as Local Brightness Correction. It has nothing to do with original beam penetration. (Actually a DRC function) 66. Changed Roles for Technique, Saving Patient Exposure 67. Changing Roles for kvp and Technique: Unlike screen film imaging, image display in digital radiography is independent of (decoupled from) image acquisition AAPM Task Group 116, 2009 Report 68. Effects on the Latent Image Vs.the Displayed Image: Everything we used to say about kvp and image contrast on a film can still be said about the subject contrast present in the remnant beam reaching the detector. In the digital age, we might consider this the latent image. By the time this information is converted into numerical data, digitized, equalized, rescaled, gradation processed, detail enhanced, noise corrected, and formatted for display, the only image quality that has not been tampered with by the computer is distortion! 69. Factors Affecting Displayed Image Contrast: 1. kvp 2. Rescaling 3. LUT applied for gradation processing 4. Reduction of local fog patterns by detail processing, 5. Windowing. Greatest impact? If kvp still controls contrast, then what do LUT s do? 70. Demo: kvp and contrast for 9 manufacturers 71. The radiologist typically begins to window the image immediately after bringing it up, according to the anatomy and the pathology to be ruled out. This renders the question of kvp and initial displayed contrast not only minor, but practically irrelevant

9 72. We must finally give up on controlling factors : When it comes to the qualities of displayed digital image, the role of the specific mas/kvp technique used is reduced to one contributing factor among many. Even though they may have been critical to making the digital image possible by ensuring adequate signal to noise ratio at the detector, stating that they control any displayed image quality adds more confusion than it alleviates 73. kvp and Mottle: Can the 15% rule be applied to reduce mas without causing significant mottle? If so, how many steps may be applied before mottle becomes significant? Study currently under way 74. At least one 15% step increase in kvp with attendant halving of mas can be applied across the board 75. Mottle Comparisons from your lab: Brightness and contrast must be equalized first 76. High kvp and Patient Dose: Starting out with a high kvp approach: Insures adequate signal penetrating through to the detector, which is critical, and Always allows some decrease in mas that results in a net savings in patient dose. To quote Dr. Hughes, Even a 10% reduction in patient dose is worth pursuing 77. High kvp Philosophy: In the digital age, to save patient dose, increases in technique should generally be made using kvp, decreases should generally be made using mas 78. Variable kvp Charts: More difficult, but worth it 79. If fixed kvp approach is used, it is still: Essential that kvp s used not fall below minimum kvp for adequate penetration for each body part, Strongly recommended that new optimum kvp s for digital equipment be adopted (a single 15% step increase from conventional techniques) 80. Issues in Applying Digital Features: 1. Understanding and use of exposure indicators 2. Dose creep 3. Latitude for radiographers in windowing and using alternate algorithms 4. Ease of surreptitiously repeating exposures 5. Resistance to change in applying high kvp tech s 6. Post annotation and legal documentation issues 7. Post collimation and other misleading concepts 8. Manufacturers use of confusing terms for various proprietary features, and focus on marketing their products rather than on patient care, (e.g. double the mas for CR ) Electronic Image Display and Quality 1. The LCD and the Nature of Pixels: Based on the polarization of light 2. Polarizing lenses use long, slender, aligned chains of iodine molecules which act as a grid. Only light waves which vibrate parallel to these chains of molecules can penetrate through Perpendicular ones are blocked 3. By placing two polarizing lenses perpendicular to each other, all light will be blocked. This arrangement is used for LCD monitor screens 4. LCD process involves tricking these layers into allowing light to pass through, by using a special material in between them which twists the light 5. This material consists of nematic liquid crystals.

10 Nematic Molecules = Have a long, thread like shape, and tend to align with each other. Liquid = able to flow around each other, even though they have crystalline structure 6. Threads of liquid crystals tend to line up with scratches on front and back electrodes (uncharged). Scratches on electrodes perpendicular to each other 7. Liquid crystals line up in spiral pattern that twists 90 degrees between the two plates Light follows the orientation of the crystals 8. Transparent conductors built into glass plates, rows in one, columns in other, act like flat shaped wires to conduct electricity. Each intersection of these constitutes a pixel 9. Normally, light twists and passes through the second polarized sheet of glass. Pixel is considered to be in on state 10. When electrical charge is applied, nematic crystals align to the charge, parallel to each other. The twisting effect is lost, light is blocked by the second polarizing filter 11. Result is dark spot in the image. When electrical charge is applied to the pixel, it is dark and considered to be shut off! Different voltages applied cause more or less twisting, resulting in various gray shades 12. To achieve adequate brightness, LCD monitors use LEDs (light emitting diodes) or fluorescent bulbs as a backlighting source. Here, a pair of thin fluorescent bulbs are mounted to the side 13. Several special filters disperse light from the side mounted fluorescent bulbs evenly across the screen 14. Rows of pixels in LCD panel form an active matrix array (AMA) 15. What really is a pixel? PIXEL = smallest screen element which can represent all gray levels (or colors) within the dynamic range of the imaging system. 16. By this definition, a single pixel on a conventional color TV screen would be considered as each triad of phosphor dots, red, yellow and blue, since combinations of these are required to cover the color spectrum 17. We are then forced to define each individual dot on the color TV as a subpixel. Some LCDs are built with a similar dot arrangement 18. In most LCDs, a single pixel actually consists of 18 individual bar like segments, three segments makes a domain, a pair of domains makes a subpixel. A pixel consists of a group of three subpixels 19. For a monochrome display, each subpixel can be treated as a separately addressed element by the computer, producing entire range of gray levels from black to white, acting as a whole pixel. We acquire 3 functional pixels in the same space that a single pixel in a color monitor requires :: 3 times higher resolution 20. LCD monitor can be visually inspected for faulty pixels and subpixels with aid of simple magnifying glass. For most LCDs, one pixel is just the size of a 12 point font period or the dot of an i. A defect smaller than this would be caused by the failure of a subpixel 21. A truly dead pixel appears as a white spot against a solid black background. A stuck pixel appears as a black spot against a solid white background. A stuck pixel is being continuously supplied with electricity

11 Time Allowing: Potpourri of Teaching Models or Display QC Potpourri: Some of my favorite models for teaching digital image production and processing: 1. Image Production: A. Visualizing Pixel Values and Detail in the Matrix: model - # digits: examples B. Illustrating Voxel Attenuation and Contrast Enhancement -Attenuation coefficients averaged throughout voxel > pixel value -Rounded by ADC to gray level # > voltage for display = Brightness -Voxels = 3D cubes for CT, but square tubes for CR & DR C. Why We Need Dels : Pixel = picture, not appropriate for physical detectors acquiring -To form dig img, info from voxels collected by dels, processed to become pixels -Dels apply to both forms of DR, Pixels to image from CR phosphor 2. Windowing: A. It has always been true that: Image brightness (or average density) can be changed without changing image contrast, and contrast (or gray scale) can be changed without changing image brightness (ave. density). They are NOT directly related. B. Wall window analogy: -Here, raising the window level, we see an overall darker image, but the range of gray shades remains equal (5, in this example) -Here, without changing the center or window level (arrow), the window width can be expanded, increasing the gray scale from 5 to 8 shades C. Controls: (CT ex): LEVEL: Controls overall or average brightness, but does NOT change the gray scale WIDTH: Lengthens the gray scale, but does NOT alter the average brightness (or average density) WL > darkness OPPOSITE Brightness WW > g scale OPP Contrast -Cut-film for window demo

Acquisition, Processing and Display

Acquisition, Processing and Display Terri L. Fauber, R.T. (R)(M) Department of Radiation Sciences School of Allied Health Professions Virginia Commonwealth University Topics Image Characteristics Image