8/3/2016

1

Digital Radiological Imaging Laboratory

Assessing Image Quality

Wei Zhao, Ph.D.

DEPARTMENT OF RADIOLOGY Breast Tomosynthesis

• Limited angular range: 11~60o

• Slice thickness: 1mm

• Total Dose: ~1 - 2 times screening mammogram

• Screening/diagnosis

• Less compression

• Fast clinical transition

COR

x-ray source

detector

projection images

breastImage slices

Different DBT System Design Company System Design View

#

Detector Recon

GE ±12.5°, step/shoot 9 CsI/a-Si, 100 um iterative

Planmed ±15º, continuous 15 CsI/a-Si, 83 um iterative

Fuji ±7.5º/±20º, continuous 15 a-Se, 68 um hex FBP

Hologic ±7.5, continuous 15 a-Se, 70 um

2x2 binning

FBP

Siemens ±25°, continuous 25 a-Se, 85 um FBP

IMS Giotto ±20°, step/shoot 13 a-Se, 85 um iterative

Philips

Microdose

±5.5º, continuous Si counting, 21 slit,

50 um

iterative

8/3/2016

2

Factors Affecting Image Quality

• Detector performance

DQE at low dose: 1/NView

Temporal performance: lag and ghosting

• Imaging system design

Focal spot blur: continuous tube travel

Geometric blur: Oblique entry of x-rays

Angular range, number of views

• Reconstruction algorithm

Analytical: FBP (filter design)

Iterative

Projection Image Quality

• Low dose: (1/N views)

Electronic noise effect

• Fast acquisition:

Lag and ghosting: Image artifacts

Pixel binning (2x2)

Detector physics: Flat panel detectors

Indirect conversion:

X-ray – light - charge

8/3/2016

3

Detector Physics: Direct conversion

Direct conversion:

X-ray - charge

Low Dose: Direct

0 1 2 3 4 5 60.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

DQ

E

Spatial Frequency (cycles/mm)

0.59 mR

1.20 mR

2.35 mR

Direct AMFPI: 85 micron pixel

Low Dose: Indirect

0 1 2 3 4 50.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

DQ

E

Spatial frequency (cycles/mm)

231 Gy

75 Gy

25 Gy

Ghetti et. al. Med. Phys. 35 (2008)

8/3/2016

4

Physics: Projection Images

• Detector performance

• Tomo acquisition cause additional blur

Moving focal spot:

• Continuous tube motion during exposure

• Gantry speed, x-ray pulse width: motion ~1 mm

Oblique entry of x-rays

Resolution: Oblique Entry of X-rays

x-ray x-ray x-ray

Detector

Blur More blurSharp

2D: Projection Image Resolution

0 2 4 6 8 10 12 140.0

0.2

0.4

0.6

0.8

1.0

MT

F

spatial frequency (cycles/mm)

oblique entry

fsm=0.65mm

detector MTF

binning fsm=1.15mm

detector MTF binning

Direct detector, 85 um

8/3/2016

5

Image Quality: 3D tomo slices

• Central slice theorem: 2D – 3D

• Angular range: Resolution, artifact

• Reconstruction algorithm: FBP, iterative

3D Tomosynthesis Sampled Space

Reconstructed slices

x

y

dxdy

dz

fz

fxz

Spatial domainFrequency domain

fzN Y

fxNY

rectangular area ( )

used for reconstruction

fz ,fxNY NY

space covered by

acquired im ages voxel

FBP: Reconstruction Filters

Reconstructed slices

x

y

dxdy

dz

fz

fxz

Spatial domainFrequency domain

fzN Y

fxNY

rectangular area ( )

used for reconstruction

fz ,fxNY NY

space covered by

acquired im ages voxel

0 1 2 3 4 5 60.0

0.5

1.0

1.5

2.0

am

plit

ud

e

Spatial frequency (cycles/mm)

Slice thickness (fz)

ramp

apodization

8/3/2016

6

3D Noise Power Spectrum

40slices

Reconstructed

3d image

Projection noise image

x

z

y

dxdy

dz

voxel

Focal spot

detector

4cm Lucite

Voxel dimension

dx=0.085 mm

dy=0.085 mm

dz=1 mm

FFT

xy plane: in-plane (reconstructed tomosynthesis images)

xz plane : in-depth

3D NPS

x

y

z

X-ray noise vs. recon

fx

fy

fz

Ramp x Hann

fz fx Ramp x Hann x

Slice thickness fy fx

In-depth In-plane

3D: Resolution and out-of-plane artifact

x

y

70 um wire tilted 20 degrees

8/3/2016

7

PSF/artifact vs. Angular Range

+20o +15o +10o +5o

3D MTF: Dependence on reconstruction filter

0

0.5 cycle/mm

0 10

x (cycle/mm)

z

SBP

RA

RA+SA

RA+SA+ST

In-plane MTF vs. Angular Range

• Increasing angular range improves MTF at low frequencies

0 2 40.0

0.5

1.0

1.5

2.0

2.5

3.0

MTF

freq(cycles/mm)

angular range

10o

20o

40o

60o

180o

8/3/2016

8

In-plane MTF

20 40 60 80 100 120 140 160 180 2000

500

1000

1500

2000

2500

3000

3500

4000

ES

F

position(pixel)

Derivative(LSF)

|1D FFT|: MTFx

Reconstructed

in-plane image

Focal spot

detector

2cm lucite

Edge phantom

Edge phantom: 0.2mm Al

In-plane MTF

Reconstructed

in-plane image

Focal spot

detector

2cm lucite

Edge phantom

Edge phantom: 0.2mm Al

0 2 4 6 8 100.0

0.2

0.4

0.6

0.8

1.0

MT

F

freq(cycles/mm)

Measurement

Model

recon setting: Rampx Hann xST, prebin

Dependence on Angular Range

±20o ±10o

ACR phantom: 28kVp, W/Rh, 1.7mGy

Recon: FBP with slice thickness filter

SDNR=2.5 SDNR=1.4

8/3/2016

9

Reconstruction filter designs

0 1 2 3 4 5 60.0

0.5

1.0

1.5

2.0

am

plit

ud

e

Spatial frequency (cycles/mm)

• Ramp + limited angle: loss of low frequency (breast

density)

• Modified filters: recovers

density, but increased out-of-

plane blur

• Iterative recon: improvement at the cost of

computation

Slice thickness (fz)

ramp

Polynomial, similar to SART

apodization

J. Zhou et. al. Med. Phys. 2007; Ludwig et. al. IWDM2008; B. Ren, et. al. SPIE 2009

Reconstruction filter comparison

FBP Iterative

Adapted from Ludwig et. al. (IWDM 2008)

FBP with enhanced low

frequency

Quality Control of DBT Systems

• Several QC protocols under development and testing

EUREF

TMIST

AAPM TG245

IEC

ACR

• Phantom design for image quality assessment in QC

8/3/2016

10

TMIST QC Phantom – middle slice

(20 mm)

BB for

ASF

Calcification

Group

NNPS Region

SDNR

sphere Calcification

Group

Calcification

Group at 10 mm

Calcification

Group at 30 mm

Courtesy: James Mainprize, Sunnybrook Research Institute

mAs and SDNR over time (Hologic)

40

45

50

55

60

65

70

mA

s

New tube

1.1

1.2

1.3

1.4

1.5

1.6

SD

NR

New tube

Unit recalibrated

Courtesy: James Mainprize, Sunnybrook Research Institute

Daily Tracking: 2D NNPS

GE Hologic Siemens

-5 -4 -3 -2 -1 0 1 2 3 4 5-5-4-3-2-1012345

f(y) mm-1

f(x)

mm

-1

-5 -4 -3 -2 -1 0 1 2 3 4 5

f(y) mm-1

-5 -4 -3 -2 -1 0 1 2 3 4 5

f(y) mm-1

Courtesy: James Mainprize, Sunnybrook Research Institute

8/3/2016

11

NNPS Tracking

Paddle Size Changes NNPS in Volume

3.00E-05

3.50E-05

4.00E-05

4.50E-05

5.00E-05

5.50E-05

NN

PS

1 c

ycle

/m

m

New Tube

Larger Paddle

Courtesy: James Mainprize, Sunnybrook Research Institute

QC Phantoms tested by TG245

• PhantomLab TomoPhan • CIRS DBT QC Phantom

Update of AAPM TG 245

WE-DE-207B-5

Wednesday at 10:55

(Room 207B)

TomoPhan: Slice Sensitivity Profiles

Courtesy: Hildur Ólafsdóttir, Image Owl

8/3/2016

12

SSP versus System Design

Courtesy: Hildur Ólafsdóttir, Image Owl

Summary

• Assessing Image Quality: Projection to 3D

2D projection images: low-dose performance

3D reconstructed slices: resolution, artifact, noise

• Impact of Imaging geometry and reconstruction

Acquisition: angular range, focal spot motion

Reconstruction: filter design

• Quality control: Phantom design and features

Monitor image quality during daily operation

Acknowledgements

• Former and current members of my lab:

Bo Zhao, Y. Hu, David Scaduto, S. Burleson, H. Huang

• Funding sources:

Siemens Healthcare

NIH 1 R01 CA148053