Lecture05 - 2009.ppt

1,012 views

Published on

  • Be the first to comment

  • Be the first to like this

Lecture05 - 2009.ppt

  1. 1. BMED-4800/ECSE-4800 Introduction to Subsurface Imaging Systems Lecture 5: X-ray Imaging (cont.) Kai E. Thomenius1 & Badri Roysam2 1 Chief Technologist, Imaging Technologies, General Electric Global Research Center 2 Professor, RensselaerPolytechnic Institute Center for Sub-Surface Imaging & Sensing
  2. 2. Review of Last Lecture • Quick historical review of X-rays was given. • Block diagrams, key components defined. • Brief discussion of x-ray scattering • An X-ray beam, traversing through an object, is attenuated by the exponential Lambert-Beer Law. • The product of the attenuation coefficient and the path length of the x-ray beam in such a target is critical in establishing detectability. • Today: – Digital Detectors, X-ray Metrics
  3. 3. Outline of Course Topics • THE BIG PICTURE – What is subsurface imaging? – Why a course on this topic? • EXAMPLE: Projection Imaging – X-Ray Imaging – Computer Tomography • COMMON FUNDAMENTALS – Propagation of waves – Interaction of waves with targets of interest • PULSE ECHO METHODS – Examples • MRI – A different sensing modality from the others – Basics of MRI • MOLECULAR IMAGING – What is it? – PET & Radionuclide Imaging • IMAGE PROCESSING & CAD
  4. 4. www.aapm.org/meetings/amos2/pdf/26-5959-83142-414.pdf
  5. 5. Digital Detector Front End www.aapm.org/meetings/amos2/pdf/26-5959-83142-414.pdf
  6. 6. Detector Details www.aapm.org/meetings/amos2/pdf/26-5959-83142-414.pdf
  7. 7. Selenium based Detector www.aapm.org/meetings/amos2/pdf/26-5959-83142-414.pdf
  8. 8. Performance Metrics
  9. 9. Signal-to-noise Ratio (SNR) • SNR determines the detectability of an object • Signal derived from x-ray quanta • Noise comes from a variety of sources: – X-ray quantum statistics, Poisson distribution – Electronic noise – Sampling noise – Anatomical noise • Signal processing steps critical to image quality – Correction for detector variability, defects – Post-process filtering FP vessel 12 mm large cell lung cancer
  10. 10. Quantum noise • For a digital x-ray detector system with square pixels – if the average number of x- rays recorded in each pixel is N, – then the noise (per pixel) will be • Statistical distribution associated with x-rays is the Poisson distribution. – The above relation falls out directly from this fact. N=σ
  11. 11. Poisson Distribution • Poisson Distribution is a probability distribution given by ( ) ( ) ! exp , k kf k λλ λ − = If the expected no. of occurrences in a space is λ, then the probability that there are exactly k occurrences is given by f(k, λ)
  12. 12. Signal-to-noise ratio • The signal-to-noise ratio (SNR) is given by • When the number of x-rays, N, is increased, the radiation dose also increases. • To double the SNR, the dose to the patient needs to be increased by a factor of 4 • Contrast-to-noise ratio (CNR) for any two intensities (I1 and I2) at a detector is given by – Here N is the nominal value of photons reaching the detector. N N NN === σ SNR CNR = I1 − I2 σ = I1 − I2 N
  13. 13. Other Measures of Image Quality • Limiting Spatial Resolution (LSR) – The highest frequency that can be visualized • Modulation Transfer Function (MTF) – Measures how the detector passes signal, as a function of spatial frequency MTF = Modulation at detector output Modulation at detector input Spatial Frequency (cycles/mm) MTF 1.0 0 0.03 - 0.05 LSR
  14. 14. MTF=1 in out Contrast Contrast MTF = in out q(x) 2A(f)) q(x) 2A(f)) Source: http://203.64.251.39/info/download/etc/breastx/93/93-04.ppt
  15. 15. MTF=0.5 • Modulation TransferFunction (MTF) /Spatial resolution: An imaging system’s ability to render the contrast of an object as a function of object detail. in out Contrast Contrast MTF = in out q(x) 2A(f)) I(x) 2Aout(f)
  16. 16. MTF
  17. 17. Modulation Transfer Function •LSR – Screen-film has LSR » 20 lp/mm • corresponds to 25 µm pixel • Digital (GE) 100 µm pixel •Sources of MTF degradation – Lateral spread of light in scintillator • limited by CsI needles • increases with scintillator thickness – Lateral spread of secondary x-rays • not significant away from k-edges of Cs and I – Sampling aperture of pixel: sinc(fx*a)sinc(fy*a) Spatial Frequency (cycles/mm) M T F Digital Imager Film-Screen If film’s LSR is better than digital why do we see improved performance in digital?
  18. 18. MTF For Direct, Indirect, and Screen Film
  19. 19. Measures of Image Quality-DQE • Detective Quantum Efficiency, DQE • SNR gives the transfer function of bothsignal and noise DQE = SNR2 at detector output SNR2 at detector input ∝ SNR2 at detector output Patient Dose
  20. 20. DQE Spatial Frequency (cycles/mm) DQE Digital Imager Film-Screen 1.0 High DQE in low-to-mid frequencies aids detection. High DQE in high frequencies aids characterization. The higher the DQE, the higher the SNR, and the greater the probability of detection.
  21. 21. where f is the spatial frequency (lp/mm), X is the exposure (mR) and : S = Median Signal Level (cts), i.e. amplitude of information MTF = Modulation Transfer Function NPS = Noise Power Spectrum (cts^2 * mm^2) C = Incident Xray Fluence (xrays / (mm^2 * mR) DQE describes the measured SNR in relation to an ideal detector. SNR2 is deduced from the ratio of MTF^2 (signal^2) to the NPS (noise^2) DQE : Definition
  22. 22. www.aapm.org/meetings/amos2/pdf/26-5959-83142-414.pdf
  23. 23. Calibration of Digital Detector • Dark Image Offset • Diode leakage • FET charge retention • Electronic noise
  24. 24. Calibration of Digital Detector • Offset Corrected Dark Image • Electronic Noise
  25. 25. Calibration of Digital Detector • Offset Calibrated • Amplifier gain variation • Pixel-to-pixel gain variation
  26. 26. Calibration of Digital Detector • Offset and Gain calibrated Flood exposed image • Poisson statistical x-ray noise • Electronic noise
  27. 27. Apply Corrections • Low dose: before and After Offset Correction
  28. 28. Apply Corrections • High dose
  29. 29. Tomosynthesis
  30. 30. Advanced Applications • Tomosynthesis- 3D X-ray
  31. 31. 3D Breast Imaging - Tomosynthesis • 3D imaging addresses the major problem with mammography today – superimposed tissue • 3D imaging may enable compression reduction – Tissue immobilization vs. compression – Compliance with screening protocols • Single tomo exam in MLO position may replace conventional mammography, potentially enabling dose reduction
  32. 32. Tomosynthesis Concept
  33. 33. Prototype System Parameters • Prototype based on GEMS Senographe DMR, Revolution flat panel detector, motorized tube motion assembly • 11 projections over +/- 25 degrees • 7.5 sec patient exam time • Total dose – 1.5x a single mammographic view – 0.75x a standard mammographic screening exam • 100 micron pixels • 1 mm (3d) slice separation
  34. 34. Tomosynthesis •Goal: • Limited 3-D reconstruction to remove overlying/underlying structure • All image planes visualized using a single acquisition •Acquisition: • Vertical tube motion • Total tube angle: 5 -15° • Numberof Projected Images: 15 – 25 • Examlength: 5 -10 sec (single breath- hold) • Slice thickness: ~1 cm • Enabled by GE Revolution™ detector: Courtesy of Duke University and Wake Forest Medical Center Rotational Axis Tube vertical motion Small Changes to Rad System allows for 3D Imaging!
  35. 35. Image Reconstruction in Tomo • Data incompleteness – From a CT perspective, data is very sparse – Limited angular range (z-resolution) – Insufficient angular sampling (streaks) – Truncated projections (inconsistency)
  36. 36. Reconstruction Concept – Shift and Add Add Reconstruction of single plane Projections at different angles Shift Vertical slice through object Reconstruction of vertical slice through object Artifacts: Out-of-plane structures appear as N low-contrast copies (N = # of projections). Contrast / “blurring” of artifacts depends on N, projection angles / tube trajectory, etc.
  37. 37. An Example… Standard 2D x- ray Images courtesy of Dr. Dan Kopans- MGH
  38. 38. Tomosynthesis – Missed Cancer Spiculated Lesion Standard Mammogram MLO Tomo Slice MLO
  39. 39. Tomosynthesis Images courtesy of Dr. Dan Kopans- MGH
  40. 40. An Example… 3D Tomosynthesi s Images courtesy of Dr. Dan Kopans- MGH
  41. 41. Rad Tomo Example Low Dose 3D Imaging!
  42. 42. Receiver Operating Characteristics
  43. 43. Receiver Operating Characteristic (ROC) curves • Most basic task of the diagnostician is to separate abnormal subjects from normal subjects • In many cases there is significant overlap in terms of the appearance of the image – Some abnormal patients have normal-looking films – Some normal patients have abnormal-looking films • ROC curves are a tool for assessing the performance of a hypothesis testing algorithms.
  44. 44. 2 x 2 Decision Matrix Actually Abnormal Actually Normal Diagnosed as Abnormal True Positive (TP) False Positive (FP) Diagnosed as Normal False Negative (FN) True Negative (TN)
  45. 45. ROC curves (cont.) • For a single threshold value and the population being studied, a single value for TP, TN, FP, and FN can be computed • The sum TP + TN + FP + FN will be equal to the total number of normals and abnormals in the study population • “True” diagnosis must be determined independently, based on biopsy confirmation, long- term patient follow-up, etc.
  46. 46. Summary • Design of digital x-ray detectors was described. • Performance metrics (MTF, DQE) for x- ray performance were given. – Justification for digital detectors was based on these. • Tomosynthesis concept introduced. • Brief review of ROC methods for hypothesis testing was given. • Next time: Introduction to CT Scanners
  47. 47. Homework • Using web resources (or sources given below), describe the key steps of the direct conversion process with amorphous Selenium. How are x- rays converted to electrons? • What is the relative performance (MTF or DQE) with respect to the CsI-Photodiode approach? • Which would you buy and why? – http://www.dondickson.co.uk/download/Challenges_of_Dire – Hoheisel et al., “Modulation transfer function of a selenium-based digital mammography system”, IEEE Proc. Nuclear Science Symposium, 2004, 3589-3593
  48. 48. InstructorContact Information Badri Roysam Professor of Electrical, Computer, & Systems Engineering Office: JEC 7010 Rensselaer Polytechnic Institute 110, 8th Street, Troy, New York 12180 Phone: (518) 276-8067 Fax: (518) 276-6261/2433 Email: roysam@ecse.rpi.edu Website: http://www.ecse.rpi.edu/~roysabm Secretary: Laraine Michaelides, JEC 7012, (518) 276 –8525, michal@rpi.edu
  49. 49. InstructorContact Information Kai E Thomenius Chief Technologist, Ultrasound & Biomedical Office: KW-C300A GE Global Research Imaging Technologies Niskayuna, New York 12309 Phone: (518) 387-7233 Fax: (518) 387-6170 Email: thomeniu@crd.ge.com, thomenius@ecse.rpi.edu Secretary: TBD

×