Tecniche Diagnostiche 3 – TC
Corso di laurea in Fisica A.A. 2002-2003
Computed Tomography Principles
• 1. Projection measurement
• 2. Scanner systems
Basic Tomographic Principle
The internal structure of an object can be reconstructed from multiple projections of the object.
Exponential Attenuation of X-ray
Ray-Sum of X-ray Attenuation
Computed Tomography Principles
• 1. Projection measurement
• 2. Scanning modes
Projection & Sinogram
Computed tomography (CT): image reconstruction from projections P(θ,t) Î f(x, y)
Computed Tomography Geometry
reconstruction matrix
focal spot detector
Matrix Representation of a Tissue Slice in CT
•
Pixel (Picture Element)•
H.U. = [µ- µ(water)/µ(water]*1000CT Display Scale
linear attenuation coefficient, µ(x,y,z) reconstructed image
displayed image
Linear Attenuation Coefficients (60 keV)
Tissue
• Brain - Grey
• Brain - White
• Cerebro-Spinal Fluid (CSF)
• Pancreas
• Liver
• Water
• Fat
µ ( χµ
−1)
• .213
• .215
• .208 - .213
• .215
• .221
• .205
• .190
Grey - White Matter Contrast
C = (.215 - .213)cm-1 * 1.0 cm
= .2 %!
CT Number µ(cm-1) 39
49
.213 .215
CT number allows the computer to present the information with a larger grey scale
Variation of Linear Attenuation Coefficients with Energy
0.14 0.16 0.18 0.2 0.22
60 70 80
Energy (keV) µ
Water Fat
Variation of H.U. with Energy
Energy (keV) 60 70 80
µ(water) .205 .193 .184
µ(fat) .190 .179 .171
H.U.(fat) -73 -73 -73
Image Display
CT Number
- Hounsfield unit
• Air: -1024
• Water: 0
• Bone: +175 to +3071 Viewing Parameters
• Window level (L)
• Window width (W)
• Zoom factor
water water
HU µ
µ µ ) = 1000 µ − (
-1024 +3071
0 255
W L
Computed Tomography Principles
• 1. Projection measurement
• 2. Scanning modes
• 3. Scanner systems
• 4. Image reconstruction
CT Scanner
Data Acquisition System (DAS)
Data Acquisition System (DAS)
First Generation
One detector Translation-rotation
Parallel-beam
160 rays x 180 views 5 minutes/per slice
First Generation CT Scanner
From Webb, Physics of Medical Imaging
Second Generation
Multiple detectors Translation-rotation Small fan-beam
Second Generation CT Scanner
From Webb, Physics of Medical Imaging
Third Generation
Multiple detectors Translation-rotation Large fan-beam
800 rays x 1000 views
<1 seconds/per slice
Ring Artifact In 3 rd Generation
Fourth Generation
Detector ring Source-rotation Large fan-beam
Fourth Generation
Detector fan eliminates ring artifacts.
3 rd and 4 th Generation
Scanners
Krestel-
Imaging Systems for Medical Diagnosis
Third & Fourth Generations
Fifth Generations
Electron-beam CT for cardiac imaging.
Electron Beam CT
Sixth Generation:
Spiral/Helical/Volumetric CT
Continuous &
Simultaneous :
• Source rotation
• Patient translation
• Data acquisition
Volume Scanning
Scan-Translate Patient, Scan-Translate Patient,
Important years in helical CT history
Single-slice 1989
Dual-slice
1992 Quad-slice
1998
Quad- Slice Single- Slice
8 times faster than single-slice
One rotation / sec
Two rotations / sec +
4 slices / rotation
Why is faster better?
•Improved temporal resolution
•Faster scanning causes less motion artifacts
•Breath holding time is reduced
•Improved spatial resolution
•Narrower collimation leads to higher resolution in the z-axis (MPR)
•Narrower collimation reduces partial volume effect
•Improved contrast media concentration
•Higher contrast media concentration due to faster infusion
•Better separation of arterial and venous phases
•Increased power (mAs)
•The widened x-ray beam and sampling of multiple slices for each rotation
allows for raised mAs
•Decreased image noise
•A direct effect of raised mAs
•Efficient x-ray tube utilization
•Faster scanning causes markedly less waiting for tube cooling
•More images from x-ray tube during tube life cycle
Pitch
Pitch = table travel (mm) per gantry rotation (1) beam collimation (mm)
Information about table travel relative to beam collimation
> 1
= 1
Pitch
Volume
Volumetric scanning Viewing
Reformatting
x y
z
Reformating multiple slices into a volume produces a volume “image” with unequal
spatial resolution in x, y, and z.
Reformatting with Interpolation
Scan a volume – view a volume
The image data from the volume can be merged into single images.
Coronal Reformat 25, 50, 100,
400, 1000 images can be reconstructed from volume.
Computed Tomography Principles
• 1. Projection measurement
• 2. Scanning modes
• 3. Scanner systems
• 4. Image reconstruction
• Uses a collimator to keep exposure to a slice
• Builds image from multiple projections
• We will assume parallel rays for now.
• Actually how first scanners worked.
• Translate and then rotate as shown in diagram
Physics of Medical Physics, EditorWebb
The 1D Fourier Transform of a projection at angle θforms a line in the 2D Fourier space of the image at the same angle.
Incident y X-rays
θ R
Incident x-rays pass through the object f(x,y) from upper left to lower right at the angle 90 + θ. For each point R, a different line integral describes the result on the function gθ(R). gθ(R) is measured by an array of detectors or a moving detector.
The thick line is described by x cos θ+ y sin θ= R
I have just drawn one thick line to show one line integral, but the diagram is general and pertains to any R.
The projection gθ(R) can thus be calculated as a set of line integrals, each at a unique R.
gθ(R) = ∫ ∫ f(x,y) δ(x cosθ+ y sin θ- R) dx dy gθ(R) = ∫ ∫ f(r, φ) δ2π ∞ (r cos (θ-φ) - R) r dr dφ
0 0
In the second equation, we have translated to polar coordinates.
Again gθ(R) is a 1D function of R.
Let’s consider the 2D FFT
F(u,v) = ∫ ∫ f(x,y) e -i 2π (ux + vy) dx dy In polar coordinates,
µ= ρcosβ ν= ρsin β
F(ρ,β) = ∫ ∫ f(x,y) exp [ -i 2π ρ(x cosβ+ y sin β)] dx dy
When x cosβ+ y sin β= constant, exp [ -i 2π ρ(x cosβ+ y sin β) is a linear phase shift. This is the Fourier transform of a shifted delta function. Let’s let the constant = R and write the complex exponential as the FT of a δfunction.
F(ρ,β) = ∫ ∫ f(x,y) F[δ( x cos B + y sin B - R)] dx dy
F(ρ,β) = ∫R∫ ∫ f(x,y) [δ( x cos B + y sin B - R)] e-i 2πρR dx dy dR
F(ρ,β) = ∫ ∫ f(x,y) ∫ [δ( x cos B + y sin B - R)] e-i 2πρR dx dy dR Recall how we wrote the projection as a double integral of f(x,y) where a delta function performs the line integral,
gθ(R) = ∫ ∫ f(x,y) δ(x cosθ+ y sin θ- R) dx dy We take the Fourier Transform of gθ(r):
F{gθ(r)}= ∫R[ ∫y∫x f(x,y) δ( x cos Β+ y sin B - R) dx dy] e-i 2πρRdR
Central Section Theorem
Again, F{gθ(r)} = F(ρ,B) and since B = θin our proof, then the F{gθ(r)} = F(ρ, θ)
So in words, the Fourier transform of a projection at angle θgives us a line in the polar Fourier space at the same angle θ.
Crude Idea 1: Take each projection and smear it back along the lines of integration it was calculated over.
Result from a back projection:
bθ(x,y) = ∫ gθ(R) δ(x cosθ+ y sin θ- R) dR
Adding up all the back projections from all the angles gives, fback-projected(x,y) = ∫ bθ(x,y) dθ
π ∞
fb(x,y) = ∫ dθ∫ gθ(R) δ(x cosθ+ y sin θ- R) dR
0 -∞
Let’s calculate an impulse response to see how the reconstruction does.
gθ(R) = δ(R)
That is, δ(x,y) causes a δ(R) projection. By calculating the back- projected image, fb(x,y), we will be calculating the impulse response.
hb(x,y) = fb(x,y) for this delta object
Recall x cos θ+ y sin θ- R = r cos (θ-φ) - R hb(r,θ) = ∫ dθ∫δ(R) δ(r cos (θ-φ) - R) dR
We can simplify the integration over R by realizing that the first delta function will be non-zero only when R=0. Then we will only have one integral across θ
= ∫πδ(r cos (θ-φ)) dθ
0
Now δ[f(x)] = ∑δ(x - xn) / |f’(xn)|
n
Only one root (zero) in range of integral
cos (θ-φ) = 0 θ-φ = π/2
θ= π/2 + φ
d r r d
h
b1 1
)] | cos(
| [
)) 2 / ( ) (
, (
0
− − +
= ∫
θ θ φ φ π
θ
φ
πδ
Back-projected impulse response
hb(r) = 1/r
fb(x,y) = f(x,y) ** 1/r
Fb(ρ,θ) = F (ρ,θ) / ρ since F{1/r}= 1/ ρ
Back projected image is blurred by convolution with 1/r
Where intuitively does the 1/r come from?
We must account for this blurring to properly reconstruct the image.
How does 1/r convolution look in image space?
In frequency space?
- Low spatial frequency data is overweighted. Filter to compensate for this. Weighted by 1/ρ.
- Solution - filter each projection by |ρ| to account for the uneven sampling density
Steps:
1) Calculate projection 2) Transform projection 3) Weight with |ρ|
4) Inverse transform 5) Back project 6) Add all angles Mathematically,
∫ dθ ∫ F -1{ F{gθ(R)} |ρ| } δ(x cosθ+ y sin θ- R) dR
The reconstruction described is known as filtered back-projection.
What is downside of the described filter, |ρ| ?
ρ
What is downside of this filter?
ρ Band limit filter to:
p0 Filter |ρ|rect (ρ/ 2ρ0)
c(R) = ?
Instead of converting to frequency domain,, use convolution in the image domain
π ∞
∫ dθ∫ [ gθ(R) * c (R) ] δ( x cosθ+ y sin θ- R) dR
0 -∞
Each projection is convolved with c (R) and then back projected.
Describe c (R) c (p) = |p|
c (R) = lim 2 (ε2- 4π2R2) / (ε2+ 4π2R2)2
ε →0
p0 - p0
c (R) = p0 [ 2 sinc 2(p0 R) - sinc 2(p0 R)]
Actual Filters : Ram-Lak : Shepp-Logan
a) Ramachandran-
Lakshminarayanan kernel
b) Shepp-Logan kernel.
Solid lines represent h(η) and circles hk