Siemens 2⋅2⋅32=128-slice

dual source cone-beam spiral CT (2005)

Basics of Clinical X-Ray

Computed Tomography EMI parallel beam scanner (1972) y

Marc Kachelrieß
z x

Institute of Medical Physics

1536 views per rotation in 0.33 s
University of Erlangen-Nürnberg 2⋅32×(672+352) 2-byte channels per view
Germany 600 MB/s data transfer rate
180 views per rotation in 300 s 5 GB data size typical
www.imp.uni-erlangen.de
2×160 positions per view

GE LightSpeed
Toshiba Aquilion
What does CT Measure?
• Polychromatic Radon transform
− dL µ ( r , E ) µ (r , E )
p( L ) = − ln dE w( E ) e

Siemens Somatom Definition with normalized detected spectrum: 1 = dE w( E )

Philips Brilliance

• Widely used monochromatic approximation:

p ( L ) ≈ dL µ ( r , Eeff )

with the effective energy being around 70 keV

Dual Source

CT-Performance (Best-of Values)

CT Basics
From Single-Slice to Cone-Beam Spiral CT
Trot collimation typ. 30 cm scan1 slices/s
• Technology 1972 300 s×42 2×13 mm --- 0.007/42
– Basic parameters
1980 2s 2 mm 20 mm, 30 s 0.5
– Detector concepts, tube technology
– Scan trajectories, scan modes 1990 1s 1 mm 10 mm, 30 s 13
• Algorithms 1995 0.75 s 1 mm 8 mm, 30 s 1.33
Spiral CT

– 2D filtered backprojection
– Spiral z-interpolation 1998 0.5 s 4×1 mm 4×1 mm, 30 s 124
– ASSR and EPBP (cone-beam recon.)
2002 0.4 s 16×0.75 mm 16×0.75 mm, 12 s 604
– Phase-correlated CT (e.g. cardiac CT)
• Image quality and dose 2004 0.33 s 64×0.5 mm 64×0.5 mm, 3 s 2404
– Spatial resolution (PSF, SSP, MTF) 2010 0.2 s 512×0.5 mm 512×0.5 mm, 0.2 s 2500
– Relation of noise, dose and resolution
– Dose values (CTDI, patient dose) 1 assuming a breath-hold limit of 30 s
– Dose reduction techniques 2 factor 4 converts from head FOM to full body FOM
3 assuming p = 1, otherwise Seff is increased
4 assuming p = 1.5 since IQ is independent of pitch for MSCT

Seite 1

1
 Fan-Beam Geometry
(transaxial / in-plane / x-y-plane)
p.a. →
x-ray tube

Acquisition

180°
x lateral →
Reconstruction
field of measurement
(FOM) and object

Object, Image
detector (typ. 1000 channels)
y
a.p. →
Sinogram, Rawdata
x
(illustration without quarter detector offset)

Data Completeness

y y

x x

In the order of 1000 projections y y

with 1000 channels are acquired Each object point must be viewed by an angular interval of
per detector slice and rotation. 180°or more. Otherwise image reconstruction is not possible.
x x
(illustration without quarter detector offset)

Basic Parameters
(best-of values typical for modern scanners) Demands on the Mechanical Design
• In-plane resolution: 0.4 … 0.7 mm
• Continuous data acquisition in spiral scanning mode
• Nominal slice thickness: S = 0.5 … 1.5 mm
• Able to withstand very fast rotation
• Effective slice thickness: Seff = 0.5 … 10 mm – Centrifugal force at 550 mm with 0.5 s: F = 9 g
• Tube (max. values): 100 kW, 140 kV, 800 mA – with 0.4 s: F = 14 g
– with 0.3 s: F = 25 g
• Effective tube current: mAseff = 10 mAs … 1000 mAs
• Mechanical accuracy better than 0.1 mm
• Rotation time: Trot = 0.33 … 0.5 s
• Compact and robust design
• Simultaneously acquired slices: M = 4 … 64
• Short installation times
• Table increment per rotation: d = 2 … 50 mm
• Long service intervals
• Pitch value: p = 0.3 … 1.5
• Low cost
• Scan speed: up to 16 cm/s
• Temporal resolution: 50 … 250 ms

Seite 2

2
 Demands on X-Ray Sources Tube Technology

conventional tube high performance tube

(rotating anode, helical wire emitter) (rotating cathode, anode + envelope, flat emitter)
• High instantaneous power levels (typ. 50-100 kW)
cooling oil cooling oil
• Increasing with rotation speed
• High continuous power levels (typ. >5 kW) cathode

anode

anode
• High cooling rates (typ. >1 MHU/minute) C B

• High tube current variation (low inertia)

• Compact and robust design C
cathode

anode
cathode anode

Photo courtesy of Siemens

Photo courtesy of GE

∂z
tan φ =
∂RF Demands on CT Detector Technology

C • Available as multi-row arrays

• Very fast sampling (typ. 300 µs)
2∂z
• Favourable temporal characteristics
(decay time < 10 µs)
• High absorption efficiency
cooling oil • High geometrical efficiency
2∂RF • High count rate (up to 108 cps*)
cathode 2∂RF
• Adequate dynamic range (at least 20 bit)
anode

B
C

* in the order of 105 counts per reading and 103 readings per second

Straton Tube

Multirow Detectors for Multi-Slice CT Adaptive Array

2006 z Technology
40 mm
GE
64 × 0.625 mm
64 / 0.37 s / 3.8°

40 mm
64 × 0.625 mm
Philips β
64 / 0.4 s / 3.8°

19 mm
16 channels
2⋅2⋅32 × 0.6 mm Siemens (of 103) shown
24 × 1.2 mm 2⋅64 / 0.33 s / 1.9°

32 mm
Toshiba
64 × 0.5 mm
M = 64 / 0.4 s / 3.2°

z
Data courtesy of Siemens Medical Solutions, Forchheim, Germany
Number of simultaneously acquired slices M / Rotation time trot / Cone-angle Γ

Seite 3

3
 Rows vs. Slices Scan Trajectories
z z z

FOM

4 Rows 6 Rows 1 Rows 4 Rows

4 Slices 4 Slices 4 Slices 8 Slices
d d 1
M = 1⋅ 4 M = 1⋅ 4 M = 4 ⋅1 M = 2⋅4 p= ≤ 1.5 p= ≤ 0.9 p=
M ⋅S M ⋅S N rot
Spiral Sequence Circle

CT Basics
From Single-Slice to Cone-Beam Spiral CT

• Technology
– Basic parameters
– Detector concepts, tube technology
– Scan trajectories, scan modes
• Algorithms
– 2D filtered backprojection
– Spiral z-interpolation
– ASSR and EPBP (cone-beam recon)
– Phase-correlated CT (e.g. cardiac CT)
• Image quality and dose
– Spatial resolution (PSF, SSP, MTF)
– Relation of noise, dose and resolution
Animation by Udo Buhl, Aachen – Dose values (CTDI, patient dose)
– Dose reduction techniques

Emission vs. Transmission 2D: In-Plane Geometry

Emission tomography
• Infinitely many sources
• No source trajectory
• Detector trajectory may be an
issue
• 3D reconstruction relatively
simple
Transmission tomography
• A single source • Decouples from longitudinal geometry
• Source trajectory is the major • Useful for many imaging tasks
issue
• Easy to understand
• Detector trajectory is an
important issue • 2D reconstruction
• 3D reconstruction extremely – Rebinning = resampling, resorting
difficult – Filtered backprojection

Seite 4

4
 Fan-beam geometry Parallel-beam geometry

transaxial
rebinning

(β ,α ) (ξ ,ϑ )

Fan-beam geometry Parallel-beam geometry

In-Plane Parallel Beam Geometry

y β y y
RF

α ϑ ξ ϑ ξ
x x x

Measurement:
(β ,α ) (ξ ,ϑ ) p(ϑ , ξ ) = Rf (ϑ , ξ ) = dx dy f ( x, y ) δ (x cos ϑ + y sin ϑ − ξ )

FBP (Filtered Backprojection) 2D Backprojection

(Discrete Version of the Transpose Radon Transform)

Measurement: p(ϑ , ξ ) = dx dy f ( x, y ) δ (x cos ϑ + y sin ϑ − ξ )

1D FT: − 2π iξu − 2π iu ( x cosϑ + y sin ϑ )

dξ p (ϑ , ξ ) e = dx dy f ( x, y ) e

Central slice theorem: P2 (ϑ , u ) = F (u cos ϑ , u sin ϑ )

π ∞
2π iu ( x cos ϑ + y sin ϑ )
Inversion: f ( x, y ) = dϑ du u P2 (ϑ , u ) e
0 −∞ p(ϑ , ξ )
π
= dϑ p(ϑ , ξ ) ∗ k (ξ )
0 ξ = x cos ϑ + y sin ϑ Add ray value to each pixel in the “vicinity“ of the ray.

Seite 5

5
 Spiral CT Scanning Principle
Filtered Backprojection (FBP) Start of Scan
trajectory
spiral scan
1. Filter projection data with the reconstruction kernel.
2. Backproject the filtered data into the image:

0 z
Direction of 0 t
continuous
Reconstruction kernels balance between patient
spatial resolution and image noise. transport
1996: 1998: 2002: 2004:
1× 5 mm, 0.75 s 4× 1 mm, 0.5 s 16× 0.75 mm, 0.42 s 2⋅⋅32×0.6 mm, 0.33 s
Kalender et al., Radiology 173(P):414 (1989) and 176:181-183 (1990)

Spiral z-Interpolation for Single-Slice CT

without z-interpolation with z-interpolation
M=1
d
p= ≤2
M ⋅S

z = zR

Spiral z-interpolation is typically a linear interpolation between points

adjacent to the reconstruction position to obtain circular scan data.

Spiral z-Filtering for Multi-Slice CT

M=2, …, 6 d CT Angiography:
p= ≤ 1.5
M ⋅S Axillo-femoral
bypass

M=4
z

120 cm in 40 s

0.5 s per rotation

z = zR ×2.5 mm collimation
4×
pitch 1.5
Spiral z-filtering is collecting data points weighted with a triangular or
trapezoidal distance weight to obtain circular scan data.

Seite 6

6
 The Cone-Beam Problem

Animation by Siemens

1×5 mm 4×1 mm 16×0.75 mm 2⋅⋅32×0.6 mm 256×0.5 mm

0.75 s 0.5 s 0.375 s 0.375 s << 1 s ?

Cone-Beam Artifacts Advanced Single-Slice Rebinning

z z z (ASSR)
3D and 4D Image Reconstruction for Small Cone Angles

• First practical solution to the cone-beam problem

Cone-angle Γ = 6°
in medical CT
• Reduction of 3D data to 2D slices
• Commercially implemented as AMPR
• ASSR is recommended for up to 64 slices
Cone-angle Γ = 14° Cone-angle Γ = 28°

Defrise phantom Do not confuse

the transmission algorithm ASSR
with
fokus trajectory the emission algorithm SSRB!

Kachelrieß M, Schaller S, Kalender WA. Med Phys 2000; 27(4):754-772

The Reconstruction Plane

For each reconstruction
z d d–Filtering in the Image Domain
position α R minimize the d
primary,
mean deviation of the R– n γ tilted imag
es
plane and the spiral
segment around α R . τ • No in-plane interpolations
R • Interpolation along d
R : n ⋅r − c = 0
• Arbitrary d-filter width
αR
sin γ cos ϕ 3 intersections
n = sin γ sin ϕ for each R-plane x, y, ξ
R
cos γ
final,
Resulting mean deviation at RF : ∆ mean ≈ 0.014 d transaxial images
'
at RM : ∆ mean ≈ 0.007 d

Kachelrieß M, Schaller S, Kalender WA. Med Phys 2000; 27(4):754-772 Kachelrieß M, Schaller S, Kalender WA. Med Phys 2000; 27(4):754-772

Seite 7

7
 Comparison to Other Approximate Algorithms Patient Images
180°LI d=1.5mm Π d=64mm MFR d=64mm ASSR d=64mm
with ASSR
• High image quality
• High performance
• Use of available
2D reconstruction
hardware
• 100% detector usage
• Arbitrary pitch

• Sensation 16 at
• 0.5 s rotation
• 16×× 0.75 mm collimation
• pitch 1.0
• 70 cm in 29 s
• 1.4 GB rawdata
• 1400 images
Bruder H, Kachelrieß M, Schaller S. SPIE Med. Imag. Conf. Proc., 3979, 2000

CT-Angiography
Sensation 64 spiral scan with 2⋅⋅32×0.6 mm and 0.375 s

CTA, Sensation 16 at

Data courtesy of Dr. Michael Lell, Erlangen, Germany

Extended Parallel Backprojection (EPBP)

Feldkamp-Type Reconstruction 3D and 4D Feldkamp-Type Image Reconstruction
for Large Cone Angles
• Approximate
volume
• Similar to 2D reconstruction:
– row-wise filtering of the rawdata • Trajectories: circle, sequence, spiral
– followed by backprojection • Scan modes: standard, phase-correlated
• True 3D volumetric • Rebinning: azimuthal + longitudinal + radial
backprojection along the • Feldkamp-type: convolution + true 3D backprojection
original ray direction
• 100% detector usage
• Compared to ASSR:
– larger cone-angles possible • Fast and efficient
ray
– lower reconstruction speed
– requires 3D backprojection hardware 3D backprojection

Kachelrieß et al., Med. Phys. 31(6): 1623-1641, 2004

Seite 8

8
 z Kymo

β The complicated
pattern of overlapping
3-fold
C data …
longitudinally
rebinned 4-fold … will become even
detector C+B more complicated with
phase-correlation.
5-fold

C Individual voxel-by-
voxel weighting and
normalization.

C: Area used for convolution

B: Area used for backprojection
Kachelrieß et al., Med. Phys. 31(6): 1623-1641, 2004 ECG

• Spiral • Spiral
• EPBP Std • EPBP Std
The 180°Condition • p = 0.375 • p = 1.0
y

ϑ
dϑ w(ϑ ) = π 180°in 3 segments

and r
x
w(ϑ + kπ ) = 1 • Spiral
k
• ASSR Std
• p = 1.0

The (weighted) contributions to each object point

must make up an interval of 180°and weight 1.
Kachelrieß et al., Med. Phys. 31(6): 1623-1641, 2004 • 256 slices
• (0/300)

EPBP Std EPBP CI, 0% K-K EPBP CI, 50% K-K

Advantages of Multi-Slice Spiral CT
• Image quality independent of scan parameters
• Increase (up to a factor of M)
– of scan speed
– of z-resolution
• New applications
– CT angiography
– dynamic studies
– virtual endoscopy
– cardiac CT
– …

Today, complete anatomical regions

are routinely scanned with MSCT
within a few seconds with isotropic
sub-millimeter spatial resolution.

Patient example, 32x0.6 mm, z-FFS, p=0.23, trot=0.375 s.

Seite 9

9
 Motion Artifacts of the Heart Cardiac CT
• Periodic motion
• Synchronisation needed (ECG, Kymogram, others)
• Prospective Gating
• Phase-correlated reconstruction = Retrospective Gating
– Single-phase (partial scan) approaches, e.g. 180°MCD
– Bi-phase approaches, e.g. ACV (Flohr et al.)
– Multi-phase Cardio Interpolation methods, e.g. 180°MCI (gold-standard)
• Generations
– Single-slice spiral CT: 180°CD, 180°CI (introduced 1996*)
– Multi-slice spiral CT: 180°MCD, 180°MCI (introduced 1998*)
– Cone-beam spiral CT: ASSR CD, ASSR CI (introduced 2000*)
– Wide cone-beam CT: EPBP (introduced 2002*)
*Med. Phys. 25(12) 1998, Med. Phys. 27(8) 2000, Proc. Fully 3D 2001, Med. Phys. 31(6) 2004

Synchronization with the Heart Phase Maximum Pitch for

teff = width / heart rate
Full Phase Selectivity
e.g. 15% / 60bpm = 150ms
Allowed data
ranges
• Voxel illumination must exceed one motion cycle
• Table increment per motion cycle must not exceed
collimation
Heart motion
width
∆c
phase p ≤ f H t rot

Sync-Signal
ECG, Kymogram, ... • E.g. trot = 0.5 s and fH = 60 bpm implies p < 0.5
R R R R • The smaller the pitch value the more segments can
0 trot 2trot 3trot 4trot 5trot 6trot t be combined
Width, and thus teff, corresponds to the FWTM of the phase contribution profile.

Kachelrieß et al., Radiology 205(P):215, (1997)

Partial Scan Reconstruction Multi-Segment Reconstruction

Table 1. Detector Combine n segments Table 1. Detector
Use one segment
position
2. Detector to obtain 180°+δδ position
2. Detector
of 180°+δδ data 3. Detector
of phase-coherent data 3. Detector
4. Detector 4. Detector
of phase-coherent data for a selected heart phase
for a selected heart phase

Time Time
Heartbeat 1 Heartbeat 2 Heartbeat 3 Heartbeat 1 Heartbeat 2 Heartbeat 3

Effective scan time Effective scan time

teff ≥ trot/2 1 teff ≥ 48 ms
Partial scan data Partial scan data
1 teff ≥ 200 ms 2 typ. 75-150 ms
(180°+ fan angle) (180°+ fan angle)
at trot = 0.4 s 3 at trot = 0.4 s

Kachelrieß, Ulzheimer, Kalender, Med. Phys. 27(8):1881-1902 (2000) Kachelrieß, Ulzheimer, Kalender, Med. Phys. 27(8):1881-1902 (2000)

Seite 10

10
 Multi-Threaded CT, Dual Source CT

Siemens SOMATOM Definition dual source

cone-beam spiral CT at the IMP

Volume Zoom, 4 × 2.5 mm, 0.5 s, 1998 Sensation 64, 2⋅⋅32 × 0.6 mm, 0.33 s, 2004
Multi-segment 180°MCI reconstruction, 90 bpm Data courtesy of Stephan Achenbach

compact
bone What is Displayed?
CT Basics
From Single-Slice to Cone-Beam Spiral CT
1000
80
800
• Technology 70
liver
600 blood
– Basic parameters spong. 60
400 bone
– Detector concepts, tube technology pancreas
CT-value / HU

50
– Scan trajectories, scan modes 200 kidney
water 40
0 fat
• Algorithms 30
– 2D filtered backprojection -200
20
– Spiral z-interpolation -400 lungs ImpactX.ocx V 1.1.0.0
10
– ASSR and EPBP (cone-beam recon.) -600
0
– Phase-correlated CT (e.g. cardiac CT) -800
air
• Image quality and dose -1000

– Spatial resolution (PSF, SSP, MTF)

– Relation of noise, dose and resolution
– Dose values (CTDI, patient dose) µ (r ) − µ Water
– Dose reduction techniques CT (r ) = ⋅1000 HU
µ Water

Spatial Resolution 1
out
In-plane resolution z-resolution
0.5 mm
0 1 in 0.4 mm
0.4 mm

out

y y z
0 1 in

out x x x

in
0 1 Std. scan, x/y UHR scan, x/y Std. or UHR scan, x/z
(0, 5000) (0, 1000) (-750, 1000)

Sensation 64, collimation: 2⋅⋅32×0.6 mm

Seite 11

11
 Spatial Resolution 3
Spatial Resolution 2 Point Spread Function (PSF), Slice Sensitivity Profile (SSP)
In-plane resolution z-resolution
Standard (no zFFS)
FWHM = 1.3 S
0.3 mm
Double z-
ImpactX.ocx V 1.1.0.0 sampling (zFFS)
FWHM = 1.0 S
0.4 mm
0.5 mm FWHM
0.6 mm

FWTM

Std. scan, x/y UHR scan, x/y Std. or UHR scan, x/z z
FWHM = Seff = effective slice thickness = freely selectable parameter during image recon.

Sensation 64, collimation: 2⋅⋅32×0.6 mm Sensation 64, collimation: 2⋅⋅32×0.6 mm

Dependencies of IQ and Dose

Tricks to Improve Resolution
• Image quality is determined by spatial resolution
• Sharp reconstruction kernels and contrast resolution (image noise)
• Lowest possible Seff • Image noise decreases with the square-root of dose
• Decrease the size of the detector pixels 1 1
σ2 ∝ ∝
• Oversampling D mAs eff
– zFFS
– αFFS • Dose increases with the fourth power of the spatial
– Detector quarter offset resolution for a given object and image noise
• Use of detector combs
eµ 2R + 1
(σ / µ ) 2 ∝
µ 2 dx 4
However, image noise becomes crucial! Noise relative to Fourth power of the
the background (=1/SNR) resolution element size

Dose Calculator
Patient Dose in CT
Typical Values for 16-Slice Scanners
Head Thorax Abdomen Pelvis

Scan range / mm 120 300 400 200

Scan time / s 0.75 0.5 0.5 0.5

Collimation / mm 16×0.75 16×0.75 16×0.75 16×0.75

Eff. mAs / mAs 320 100 160 160

Critical organ Brain Lung Stomach Colon

Organ dose / mSv 54.6 11.2 15.7 11.7

Eff. dose / mSv 2.9 3.8 8.2 3.9

Eff. dose / 2.1 mSv 1.4 1.8 3.9 1.9

Routine protocols, 120 kV, male phantom.

Demo version of ImpactDose available at www.vamp-gmbh.de

Seite 12

12
 Strategies for Dose Reduction Standard Display
Potential reasons for an increase:
• Higher volume coverage
• Multiphasic examinations
• More examinations
• Higher spatial resolution
• New special applications

Potential ways to decrease dose:

• New display techniques
• Advanced reconstruction (MAF)
0,5×0,5×0,5 mm3
• Automatic exposure control (AEC) C = 50 HU, W = 400 HU
• Optimized spectra
Inside Story (Overexposure), Elvgren, 1959
• Dose training (dose tutor)

Sliding Thin Slab (STS) Display Tube Current Modulation

1 000 000

500 1 000 000

(attenuation: 2000)

0,5×0,5×10 mm3 20 000

C = 50 HU, W = 400 HU (attenuation: 50)

Constant tube current: High, inhomogeneous noise.

2 2
σ pixel = const. ⋅ σ projection ,n
n

Tube Current Modulation Dose Reduction by Tube Current Modulation

Rule of thumb:
250 000 The number of quanta reaching
the center of the patient should
be constant for all view angles.

875 1 750 000

(attenuation: 2000)

Constant
5 000
(attenuation: 50)
total mAs! Conventional scan: 327 mAs Online current modulation: 166 mAs
53% dose reduction on average for the shoulder region
Modulated tube current: Low, homogeneous noise. 49% dose reduction in this case
2 2
σ pixel = const. ⋅ σ projection ,n Kalender WA et al. Med Phys 1999; 26(11):2248-2253
n

Seite 13

13
 Automatic Exposure Control (AEC) Multidimensional transaxial filtering
(z-dependent + angular dependent tube current modulation) Adaptive Filtering
a)
(MAF)
Standard CT AEC • Rawdata based
• Local smoothing of noisy data
rel. tube current

rel. tube current

rel. image noise

rel. image noise

(less than 5% modification)
α b)
• No loss of spatial resolution
• Efficient
β
• Noise reduction can be
equivalently converted to dose
reduction
a) Low attenuation: Filter width = 0
b) High attenuation: Filter width > 0
z C 40/W 500 z C 40/W 500 pMAF ( β , α , b) =
34% mAs reduction with AEC at constant image quality for that specific case dβ ′ dα ′ db′ f ∆β ( β − β ′) f ∆α (α − α ′) f ∆b (b − b′) p ( β ′, α ′, b′)
Kachelrieß M, Watzke O, Kalender WA. Med Phys 2001; 28:475-490

Standard 180°MFI

Standard 180°MFI Noise image (standard 180°MFI)

100%
1% data modified
Adaptive 180°MAF

Adaptive 180°MAF Noise image (adaptive 180°MAF)

180°MAF relative to 180°MFI: 51%

Noise Dose Resolution Difference images

left: 61% 37% 97%
center: 63% 40% 97%
right: 60% 36% 97%
Difference image upper: 100% 100% 100%

Noise in the shoulder region typically reduced to 50%...70%.

collimation 4×1 mm, d = 5 mm, (C=0 / W=500) collimation 4×1 mm, d = 5 mm, (C=0 / W=500)

Summary
• CT technology is further evolving towards
Thank You!
– more slices
– faster rotation times
– higher spatial resolution
• CT algorithms
– reconstruct cone-beam data for any trajectory
– perform phase-correlated imaging (4D)
– significantly reduce artifacts (beam hardening, truncation …)
• CT dose
– is becoming more and more an important issue (also in the US)
– is being reduced by manufacturers‘ efforts (e.g. MAF, AEC)
3d .org
– can be most significantly reduced by user training www.fu lly

Seite 14

14