8/3/2016

1

Low-Dose Cone-Beam Breast CT:

Physics and Technology Development

John M. Boone Professor of Radiology and Biomedical Engineering

University of California Davis School of Medicine Sacramento, California

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

2

Corporate Disclosures (required by UC Davis): CONSULTANT

Canadian Association of Radiologists

RESEARCH FUNDING

Siemens Medical Systems

Varian Imaging Systems

Stanford Research Institute

ROYALTIES

Samsung Corporation (patent)

Lippincott Williams and Wilkins (book)

TRAVEL FUNDING

American Association of Physicists in Medicine (AAPM)

Japan Society of Medical Physics

OTHER CONFLICTS

Patents Pending on various breast CT concepts

California BCRP 7EB-0075 California BCRP 11I-0114 R01 CA•89260 R01 EB•002138-10 (BRP) R01 CA•129561 (RDB) P30 CA•093373 (CCSG) Susan G. Komen Foundation University of Pittsburgh Varian Imaging Systems UC Davis Bridge Funding California BCRP 20IB-0125 (Merced) R01 CA●181081

Funding Acknowledgements:

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Introduction Breast CT Technology Design & Fabrication & Integration Acquisition Performance Metrics Breast CT Clinical Trials Anatomical Noise: Computer Observer Performance Human Observer Performance Understanding Breast Anatomy (Mining the Breast CT data) Breast Density (amplitude) Breast Density (distribution) Breast Volume and Shape Summary and Conclusions

8/3/2016

2

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Introduction Breast CT Technology Design & Fabrication & Integration Acquisition Performance Metrics Breast CT Clinical Trials Anatomical Noise: Computer Observer Performance Human Observer Performance Understanding Breast Anatomy (Mining the Breast CT data) Breast Density (amplitude) Breast Density (distribution) Breast Volume and Shape Summary and Conclusions

Jemal A, et al., Cancer Statistics 2006

dea

th r

ate

per

10

0,0

00

1930 2002

Cancer Mortality and Screening

5

Breast

6

detector

0.50 mm3

0.0093 mm3

54 X

Dedicated Breast CT Mammography

8/3/2016

3

7

NPSa(f) ≈ a f -b

texture

Signal Background

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Introduction Breast CT Technology Design & Fabrication & Integration Acquisition Performance Metrics Breast CT Clinical Trials Anatomical Noise: Computer Observer Performance Human Observer Performance Understanding Breast Anatomy (Mining the Breast CT data) Breast Density (amplitude) Breast Density (distribution) Breast Volume and Shape Summary and Conclusions

Tony

Seibert

Karen

Lindfors

John

Boone

Simon

Cherry

Katie

Metheany

Shonket

Ray

Nathan

Packard

Dandan

Zheng

Ramsey

Badawi

George Burkett

Whit Miller

Fareedah Simon

John Brock

Clare

Huang

Kai

Yang

Orlando

Velazquez

Hong

Zhou

Alex

Kwan

Jessie

Xia

Carey Floyd

Tom

Nelson

Craig

Abbey

Norbert

Pelc

Bruce

Hasegawa

Elizabeth

Krupinski

Anita

Nosratieh

Nicolas Prionas

Sarah McKenny

Lin Chen

Loren Niklason

Jeff Siewerdsen

Martin Yaffe

Linda Phelps

Laurie Boling

Shadi

Shakeri

Andrew Hernandez

Peymon Gazi

Heather Johnson

Desirẻe Lazo

8/3/2016

4

Design, Fabrication and Integration

10 Images from PIXABAY.com (royalty free images)

Computer aided design / computer aided manufacture (CAD/CAM)

Cambria 2011

Bodega 2007

Albion 2003

Doheny 2014 11

12 Gantry Views System views

George Burkett

Doheny: Design

8/3/2016

5

13

Components

Varian TFT Dexela CMOS

Kollmorgen Yaskawa

Comet X-ray Varian 1501 X-ray

Motor / bearing / encoder

Detectors

X-ray tubes

Albion 2004 Bodega 2007

Doheny 2014 Cambria 2012

15

Doheny: Fabrication

8/3/2016

6

Wiring

16

before after

Gantry motor / bearing & angle encoder

Flat panel detector

PET head 1

PET head 1 actuator

X-ray tube

On-gantry computer

PET head 2 actuator

PET head 2

Filter changer stepper motor

Collimator changer stepper motor

Heat exchanger

Software: Hardware Integration

17

X-ray tube vertical actuator

console computer recon computer

X-ray generator

DOHENY

Peymon Gazi

Breast Immobilization & Beam Equalization

Andrew Hernandez

8/3/2016

7

19

Six phantoms (V1-V6)

1st 2nd

N = 219 : ~ 5 sets of 43

Largest 5 breasts

3rd 4th 5th

Mean volume and shape in each quintile

hot water bath molding breast immobilizer Aquaplast® thermoplastic

20

• Physical Dosimetry

• Image Quality Assessment

• Mold for breast immobilization

21

Breast Alignment System

Breast Containment and Alignment

mold

8/3/2016

8

Notes: V3 Phantom

Beam Shaping Filter

NO FILTER

3D BMF

20,000

15,000

10,000

5,000

0

2D-filter

detector

Acquisition Details

• 500 views are acquired

• 16.6 seconds @ 30 FPS, or 10 seconds @ 50 FPS

• 2 x 2 binning of detector elements

• Varian TFT Panel: 1024 x 768 (388 mm) at 30 FPS

• Dexela CMOS Panel: 1933 x 1533 (150 mm) at 50 FPS

• FBP Recon to 5123 or 10242 x 512

Performance Metrics Spatial Resolution

24

MTF

Spatial Frequency (mm-1)

PSF(x,y) Image a 70 mm wire

( ) ( , )LSF x PSF x y dy

2( ) ( ) ifxMTF f dx LSF x e

8/3/2016

9

25

Engineering impacts resolution

Center of FOV

Edge of FOV

Bodega

Bodega

26

Engineering impacts resolution

Center of FOV

Edge of FOV

Cambria

Cambria

pulsed acquisition (4 ms)

27

Engineering impacts resolution

Center of FOV

Edge of FOV

Doheny

Doheny

smaller dexels

pulsed acquisition (4 ms) + higher resolution detector

8/3/2016

10

Performance Metrics Contrast Resolution

28

total noise

(variance) anatomical noise

quantum noise

NPS(f) = NPSa(f) + NPSq(f)

quantum noise

29 cone angle

coronal sagittal

Contrast Resolution: NPS measurements

axial

30

Slice Thickness

x-ray tube mA

RECON algorithm

Shepp-Logan

RECON algorithm

Ramp

Near-Neigh

Bilinear

Yang et al., Noise power properties of a cone beam CT system for breast cancer detection, Med Phys. 2008

Noise Power Spectrum (NPS) measurements (Bodega)

8/3/2016

11

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Introduction Breast CT Technology Design & Fabrication & Integration Acquisition Performance Metrics Breast CT Clinical Trials Anatomical Noise: Computer Observer Performance Human Observer Performance Understanding Breast Anatomy (Mining the Breast CT data) Breast Density (amplitude) Breast Density (distribution) Breast Volume and Shape Summary and Conclusions

Clinical Trials

• Over 600 women on UC Davis scanners

• women with suspicion of breast cancer (BIRADS 4 & 5)

• Informed consent / HIPAA compliant

• 10 - 16 second scan (breath hold)

• >200 have had contrast injection

32

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Breast CT Clinical Trials

Images

Anatomical Noise

Computer Observer Performance

Human Observer Performance

8/3/2016

12

Before Patient Imaging

34

Radiation Dose

35

True 3D Display !

Nathan Packard

36

296

second volunteer imaged: January 2005

8/3/2016

13

37 first breast cancer imaged: January 2005

38

bCT (no injected contrast)

39

PRE CONTRAST POST CONTRAST

PRE CONTRAST POST CONTRAST

bCT (with contrast)

8/3/2016

14

40 Pt 160

Contrasted Enhanced breast CT pre

pre post

post

Malignant

benign

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Breast CT Clinical Trials

Images

Anatomical Noise

Computer Observer Performance

Human Observer Performance

42

total noise anatomical noise

quantum noise

NPS(f) = NPSa(f) + NPSq(f)

Burgess, et al (2001)

NPSa(f) = a f -b

anatomical noise

Anatomical Noise

8/3/2016

15

43

mammo

tomo

Breast CT

Hanning filter 2D-NPS

S/n

NPSa(f) = a f -b

Na(f)

Nq(f)

b

NPS(f) = NPSa(f) + NPSq(f)

NPSa(f) = a f -b

b slope

44

Breast CT, Tomosynthesis, and Mammography Texture Comparisons

breast CT mammography tomosynthesis

N = 23 pts 1000 ROIs per image type

45

8/3/2016

16

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Breast CT Clinical Trials

Images

Anatomical Noise

Computer Observer Performance

Human Observer Performance

Computer (PWMF) Observer Performance

47

Real breast CT data sets (N=151) Simulated Spherical Lesions from 1 mm to 15 in diameter

Signal Known Exactly (SKE)

Evaluated versus slice thickness (from 0.4 mm to 44 mm)

bCT “mammo”

Simulated lesion insertion into real breast CT data sets with different slice thickness

DI

Modulation (blurring)

Lesion Intensity

no lesion

other lesion insertion models

our model

adaptive lesion insertion model

8/3/2016

17

49

FT{S(x,y)}

PS(fx,fy) PWMF = FT-1

Pre-whitened Matched Filter

mammography

breast CT

S(x,y): 1000 samples

PS(x,y): 1000 samples

For 9 slice thicknesses

From 0.3 to 44 mm

50

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1 - specif icity

sens

itivi

ty

For each bCT data set:

1000 true lesions &

1000 non-lesions

SKE lesion PWMF template applied

AUC

S u

1 - specificity

sen

siti

vity

95% specificity

sensitivity

AUC

0

20

40

60

80

100

120

0 10 20 30 40 50 60 70 80 90

Relative Filter Response (%)

Freq

uenc

y

non-lesions

lesions

no yes

1 - specificity

sen

siti

vity

u

nu

mb

er o

f le

sio

ns

over all pixels

51

Computer (PWMF) Observer Performance

Breast CT (0.3 mm)

“Mammography” (44 mm)

(11 mm)

40% D in Sensitivity @ 3 mm

16% D in Sensitivity @ 10 mm

8/3/2016

18

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Breast CT Clinical Trials

Images

Anatomical Noise

Computer Observer Performance

Human Observer Performance

Human Observer Performance

53

(a) (b)

2-Alternative Forced Choice Design

CT images

Human Observer Performance

54

(a) (b)

2-Alternative Forced Choice Design

Projection images

8/3/2016

19

55

Computer (PWMF) Observer

Radiologist Observers*

Thickness = 0.3 mm (breast CT)

Thickness = 44 mm (“mammography”)

*average of 3 fellowship-trained (in breast imaging) radiologists

Ob

serv

er P

erfo

rman

ce (

Az)

2010

DHU AUC = 0.87

57

AUC = 0.87

Shadi Shakeri Karen Lindfors

AUC = 0.95

8/3/2016

20

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Introduction Breast CT Technology Design & Fabrication & Integration Acquisition Performance Metrics Breast CT Clinical Trials Anatomical Noise: Computer Observer Performance Human Observer Performance Understanding Breast Anatomy (Mining the Breast CT data) Breast Density (amplitude) Breast Density (distribution) Breast Volume and Shape Summary and Conclusions

59

Breast Density (amplitude)

segmentation

validation

60

N = 2831

Median VBD = 16%

Breast Density (amplitude)

50% / 50%

8/3/2016

21

61

Breast Density (distribution)

62

Breast Density (distribution)

Original Mammo

Threshold T2

Threshold T3

From Breast CT segmentation

From thresholded mammogram

50% GF on mammo

16% GF on bCT

Huang S.-Y., Boone J.M. et al. “The characterization of breast anatomical metrics using dedicated breast CT” Med Phys. 38 (4), 2011

r=0 r=1

RG

F1

(%)

0

20

40

60

80

100

0.0 0.2 0.4 0.6 0.8 1.0

0

20

40

60

80

100

0.0 0.2 0.4 0.6 0.8 1.0

0

20

40

60

80

100

0.0 0.2 0.4 0.6 0.8 1.0

Relative radial distance (r)

RG

F2

(%)

RG

F3

(%)

Relative radial distance (r)

Relative radial distance (r)

Breast Density (distribution)

Radial glandular fraction

8/3/2016

22

64

Breast Volume and Shape

Molds for Breast Immobilization

Physical Dosimetry (polyethylene ~ adipose)

Image Quality Assessment Protocol Optimization

Low-Dose Cone-Beam Breast CT: Physics and Technology Development

Advances in Dedicated Breast CT AAPM Annual Meeting 2016

Introduction Breast CT Technology Design & Fabrication & Integration Acquisition Performance Metrics Breast CT Clinical Trials Anatomical Noise: Computer Observer Performance Human Observer Performance Understanding Breast Anatomy (Mining the Breast CT data) Breast Density (amplitude) Breast Density (distribution) Breast Volume and Shape Summary and Conclusions

66

Summary:

• Breast CT has superior mass detection than mammography, based upon texture analysis, computer and human observer studies

• CE breast CT highlights malignant calcifications and is likely equivalent to CE-breast MRI

• Breast CT is FDA approved for diagnostic breast imaging, need to push the technology to achieve superior screening performance

• Breast CT is an emerging technology which will have an important role in reducing breast cancer mortality in the near future.

8/3/2016

23

University of California Davis

Breast Tomography Project