Koeppe ppt

Basic Principles and Controversies in PET Amyloid Imaging

Robert A. Koeppe, Ph.D.University of Michigan

January 13, 2012

TOPICS

• Reference Tissue Considerations- theoretical vs. practical issues

• Target Region Considerations- Region selection and VOI definition

• Thresholds for amyloid positivity- tracer, reference tissue, data extraction method, scanner model

dependent

• [18F]AV-45 vs. [11C]PiB

• Partial Volume Correction

Reference

Tissue

Considerations

• While cerebellar gray matter may be the best region to use as a reference in terms of it being the gray matter tissue with the least amount of amyloid in binding Alzheimer’s-type dementia, practical reasons, such as the low tracer retention and high dependence on accurate scatter and attenuation corrections, raise the question whether it is the best choice as the normalizing region in either longitudinal or cross-sectional studies.

• Pons and white matter have higher uptake, hence a more robust signal, but whether the uptake levels in these structures vary across subjects or differ with disease are pertinent questions. Furthermore, determining a white matter region can be difficult.

• The ability to accurately identify amyloid “positive” from amyloid “negative” (quantitatively) is dependent more on the ability to accurately determine the reference tissue value than the target tissue values. (SUVr goes up when the target increases OR when the reference decreases)

Reference

Tissue

Considerations

• Five different reference measures were evaluated:

1) Original images from ADNI (atlas VOI of cereb gray)

2) Cerebellar gray (defined on subj’s co-reg’d FDG or MR)

3) Pons (defined on co-registered FDG or MR)

4) White Matter (defined on AV45 or co-registered MR)

5) Combined (cerebellum, vermis, white matter, pons)

Target

Region

Considerations

• There are two primary questions for assessing target regions:

1) Which regions should be examined?

2) How best can the amyloid measures be extracted?

• Which regions? Standard regions defined from previous work. For this presentation, target regions included: posterior cingulate, superior parietal, lateral frontal, medial frontal, lateral temporal, and occipital cortex, and in addition the averages of all 6 or just the first 5.

• If the focus involves using PET as a biomarker for clinical trials, regions defined in this manner (i.e. pre-defined based on the target population) should be sufficient.

• However, if the focus is on early detection with the binary outcome (+ / -), means of examining the entire brain may be beneficial.

Target

Region

Considerations

• There are two primary questions for assessing target regions:

1) Which regions should be examined?

2) How best can the amyloid measures be extracted?

• How best to extract? Since in normal amyloid scans, cortical gray matter values are lower than white matter, standard data extraction methods not be optimal.

• Four different methods were evaluated:

• atlas-based VOIs determined directly from amyloid images

• VOIs defined on FDG images, then sampled from co-registered amyloid images

• VOIs defined on segmented-MR images, then sampled from co- registered amyloid images

• Surface maps of the brain defined from “peak” values in either FDG or segmented MR

Thresholds for

Amyloid

Positivity

• For PET studies where the focus is on early detection the primary question becomes, what is the threshold for being amyloid positive? This optimal threshold is dependent on:

• Radiotracer: F-18 tracers will not be the same as PiB

• Choice of reference region: Causes large differences in the threshold, but these are easily adjusted for based on the relative group means of the different reference tissues

• Target regions selected: (not strongly, however)

• Data extraction method: (for both target and reference regions)

• Scanner model and software version:

• Sensitivity / specificity trade-off

Analysis of

ADNI

AV45 scans

• Using AV45 scans from ADNI, we can examine the following:

1) Reference regions*: cerebellum, pons, white matter, combined

2) Target regions*: posterior cingulate, superior parietal, lateral frontal, medial frontal, lateral temporal, and occipital cortex

• Scanner model and software version:

1) Subjects: Mostly controls, EMCI and MCI1)Controls (n = 138)2)EMCI / MCI (n = 305) 3)AD (n = 19)

Scan data: 20 min imaging (50-70 min post bolus injection)

* All regions defined on co-registered FDG except white matter (defined on AV) then applied to unsmoothed AV45 scans

AV45 scans:

MCI subjects (n

= 305)

Cerebellar

Gray;

Pons;

White

Matter;

Cortex (5 region avg.)

AV45 scans:

MCI subjects (n

= 305) (sorted)

Cerebellar

Gray;

Pons;

White

Matter;


AV45 scans:

NC subjects (n

= 138) (sorted)

Cerebellar

Gray;

Pons;

White

Matter;


Cerebellar

Gray;

Pons;

White

Matter;


AV45 scans:

MCI subjects (n

= 305)

Amyloid

positivity threshold –

Cerebellar

Gray ref. region

Amyloid


Pons reference region

Amyloid


White

Matter ref. region

Amyloid


Combined reference region

Mean (COV) and Positivity Threshold*

* Positivity threshold calculated from the 100 controls (of 138) with lowest mean cortical amyloid values. Threshold = region mean + 3 SD

Tissue Concentration Ratio Differences between Amyloid Negative NC and MCI Subjects

Amyloid Positivity by Target Region

Group# of Positive for 4 ref reg approaches

Lateral Frontal

Medial Frontal

Posterior Cingulate

Superior Parietal

Lateral Temporal Occipital

NC138

0 (none) 69.6% 68.8% 65.9% 66.7% 68.1% 73.9%

1 2.9% 2.2% 5.1% 3.6% 2.9% 6.5%

2 2.2% 2.2% 5.1% 3.6% 4.3% 3.6%

3 11.6% 9.4% 7.2% 7.2% 6.5% 4.3%

4 (all) 13.8% 17.4% 16.7% 18.8% 18.1% 11.6%

MCI 305

0 (none) 48.9% 48.9% 48.5% 49.5% 51.1% 56.1%

1 4.9% 5.9% 7.5% 4.3% 3.9% 10.2%

2 2.3% 2.0% 2.0% 1.0% 2.3% 3.9%

3 13.4% 6.6% 8.9% 8.5% 9.8% 9.2%

4 (all) 29.5% 36.6% 33.1% 36.7% 32.8% 20.7%

Amyloid Positivity by Reference Region

Group NC (n=138) MCI (n=305)

# Target Regions Positive

CerebGray Pons White

MatterCombin

ationCereb Gray Pons White

MatterCombination

0 (none) 68.1% 68.8% 67.4% 65.9% 57.0% 54.4% 42.3% 51.8%

1 2.2% 3.6% 2.2% 1.4% 2.3% 1.6% 5.6% 1.3%

2 3.6% 4.3% 3.6% 4.3% 3.0% 2.6% 2.0% 1.6%

3 2.9% 2.2% 0.7% 2.2% 1.6% 1.0% 1.3% 1.6%

4 7.2% 2.9% 2.9% 4.3% 5.6% 3.6% 3.3% 2.3%

5 3.6% 3.6% 5.8% 5.8% 11.5% 10.2% 6.6% 10.5%

6 (all) 12.3% 14.5% 17.4% 15.9% 19.0% 26.6% 39.0% 30.8%

• Select 261 amyloid negative subjects: 100 Controls; 161 EMCI/MCI

• Scanner Models: 3 vendors; 13 models

- GE (n = 64); All Discovery PET/CT; (STE, ST, LS, RX)

- Philips (n = 39); All Gemini PET/CT; 34 TF model; 5 older models

- Siemens PET/CT (n = 78); 71 TruePoint/mCT; 7 older BioGraphs

- Siemens PET only (n = 59); 56 HR+, 3 Accel

- Siemens HRRT (n = 6); older scatter/atn correction software

- Siemens HRRT (n = 15); newer scatter/atn correction software

Scanner/Software

Effects: ‘

Negative’

NC and

MCI

Subjects

Scanner/Software Effects: ‘Negative’ NC and MCI Subjects

[18F]AV45 vs. [11C]PiB

• It’s NOT about higher white matter values in AV45

• It’s about lower cortical signal in AV45

• but it’s also about the lower noise in AV45 (important for visual reads)

• Improved reading of AV45 scans may be helped by surface maps of cortical amyloid signal. Aids in visualizing the spatial pattern of amyloid.

• 29 subjects from ADNI having both PiB & AV45

• AV45 always acquired after PiB (1 yr, 22; 2 yrs, 6; 3 yrs, 1)

• Subjects divided into amyloid negative (12) and positive (17)

• All regions (target and reference) defined on co-registered MR

• Standardized image display scale used- Display maximum is 2.5 times the difference in mean cortical SUVr

between amyloid positive and amyloid negative groups above the negative group mean

- Display maximums calculated for both tracers and each reference tissue

- (if mean cortical binding = 1.8 in the amyloid positive group and 1.2 in the negative group, then display max = 2.5 (1.8 - 1.2) + 1.2 = 2.7

[18F]AV45 vs. [11C]PiB

AV – negativeCerebellar Gray Normalization

White Matter Normalization

Pons Normalization

Combined Normalization

PiB – negativeCerebellar Gray Normalization


Pons Normalization


AV – positiveCerebellar Gray Normalization


Pons Normalization


PiB – positiveCerebellar Gray Normalization


Pons Normalization


Amyloid Surface Map Creation

• Co-registration of amyloid and FDG image sets to MR

• FreeSurfer Atlas VOIs onto individuals MR scan

• Assign fixed intensities to gray matter, white matter, CSF and sub-cortical regions from FreeSurfer segmented MR

• Smooth segmented MR to PET resolution

• Use smoothed segmented MR to determine PET voxels with the purest gray matter signal (i.e. least spillover from white matter and CSF) for creation of surface maps (NeuroStat)

• Create surface maps of co-registered amyloid images using voxel locations from smoothed-segmented MR

MR segmentation (FreeSurfer)

Co-registered AV45 PET


MR segmentation smoothed to PET resolution

Smoothed/segmented MR oriented to

standardized PET

AV45

Final Surface Maps

PiB

Pre-defined surface regions:

Peak surface pixels located on smoothed

MR then applied to

amyloid images

Pre-defined regions in

surface map orientation

AmyloidNegative

Cerebellar Gray Normalization


Pons Normalization


AV

PiB

AV

PiB

AV

PiB

AV

PiB

AmyloidPositive

Cerebellar Gray Normalization


Pons Normalization


AV

PiB

AV

PiB

AV

PiB

AV

PiB

AV45

AmyloidNeg/Pos ?

PiB

Amyloid Neg/Pos?

AV45

PiB

Amyloid Neg/Pos ?

AV45

Group average images

AV – negative (n=12) row 1 AV – positive (n=17) row 2

PiB – negative (n=12) row 3PiB – positive (n=17) row 4

Group average surface maps AV – negative row 1 AV – positive row 2PiB – negative row 3 PiB – positive row 4

The relative values of

cerebellar gray, pons and white

matter are similar for PiB

and AV45 (slopes ~1.0 across these structures).

Slopes for cortical regions are ~0.60-0.65

(for all reference

tissue normalizations

)

Cortical VOI magnitudes for AV45 and PiB (+group: mean ± SD)

Fractional Signal Above Cerebellar Gray

Reference Region Tracer

Posterior Cingulate

Superior Parietal

Lateral Frontal

Medial Frontal


Cerebellar Gray

AV45

PiB

0.45±0.33

0.72±0.65

0.38±0.43

0.74±0.65

0.36±0.37

0.61±0.53

0.34±0.33

0.58±0.48

0.31±0.33

0.53±0.49

0.30±0.29

0.44±0.35

Ratio 62.0% 51.9% 59.1% 58.2% 58.1% 67.9%

Pons AV45

PiB

0.44±0.37

0.74±0.54

0.38±0.46

0.77±0.78

0.36±0.41

0.64±0.66

0.33±0.37

0.61±0.60

0.30±0.38

0.55±0.61

0.29±0.33

0.45±0.43

Ratio 59.5% 48.9% 55.8% 54.9% 54.9% 65.5%

White Matter AV45

PiB

0.42±0.26

0.71±0.44

0.36±0.38

0.74±0.70

0.34±0.33

0.61±0.58

0.32±0.29

0.59±0.53

0.29±0.29

0.53±0.54

0.29±0.30

0.44±0.39

Ratio 59.8% 48.8% 55.2% 54.2% 54.7% 65.4%

Combined AV45

PiB

0.43±0.28

0.72±0.44

0.37±0.39

0.75±0.68

0.35±0.34

0.62±0.56

0.33±0.29

0.59±0.51

0.30±0.31

0.53±0.52

0.29±0.27

0.44±0.36

Ratio 60.0% 49.7% 56.7% 55.7% 55.9% 66.3%

Mean Z-scores for ‘High’ Individuals relative to ‘Low’ Group

Reference Region

Target VOI Method

Lateral Temporal AV / PiB

Superior Parietal AV

/ PiB

Medial Frontal AV / PiB

Lateral Frontal AV / PiB

Posterior Cingulate AV / PiB

Occipital AV / PiB

Cereb Gray Atlas VOIs 3.8 / 8.5 3.7 / 7.2 3.6 / 8.8 3.2 / 7.5 2.1 / 4.8 2.7 / 3.9

FDG mask 8.1 / 12.2 7.5 / 9.6 5.6 / 9.3 5.4 / 8.6 3.6 / 8.1 4.9 / 7.1

MR mask 7.6 / 12.3 9.8 / 6.7 6.4 / 7.6 7.1 / 9.3 6.6 / 5.6 4.4 / 4.4

Pons Atlas VOIs 3.1 / 5.8 3.3 / 5.6 3.4 / 5.4 2.7 / 4.9 1.9 / 5.1 2.9 / 3.7

FDG mask 4.9 / 8.5 4.9 / 6.9 5.0 / 7.2 4.1 / 6.5 2.9 / 5.1 2.8 / 4.7

MR mask 6.1 / 12.4 5.1 / 8.4 5.5 / 8.2 4.9 / 8.6 4.0 / 7.2 3.5 / 5.3

White Matter Atlas VOIs 8.9 / 10.0 9.7 / 14.9 7.4 / 8.7 5.1 / 6.9 2.9 / 9.2 4.7 / 8.1

FDG mask 9.2 / 11.9 8.8 / 11.7 8.5 / 8.3 7.1 / 7.4 3.8 / 5.6 4.8 / 7.9

MR mask 7.3 / 9.9 7.7 / 14.0 8.6 / 9.5 6.8 / 9.0 6.0 / 11.1 4.1 / 7.4

Combined Atlas VOIs 6.7 / 10.9 6.3 / 9.8 6.0 / 9.2 4.9 / 7.8 3.1 / 8.6 4.1 / 6.6

FDG mask 10.6/13.4 9.6 / 10.5 7.9 / 9.1 7.1 / 8.4 4.1 / 7.4 5.2 / 8.7

MR mask 11.0/15.6 11.0/11.9 11.9/11.7 8.2 / 11.4 7.3 / 10.6 4.6 / 7.4

Zindividual = [ SUVrind – mean(SUVrlo) ] / SD(SUVrlo)

Summary Z-scores Measures

Lateral Temporal AV / PiB

Superior Parietal AV / PiB

Medial Frontal AV / PiB

Lateral Frontal AV / PiB

Posterior Cingulate AV / PiB

Occipital AV / PiB

7.28 / 10.95 7.28 / 9.77 6.65 / 8.58 5.55 / 8.03 4.03 / 7.37 4.06 / 6.27

6 / 9 2 / 2 3 / 1 1 / 0 0 / 0 0 / 0

Target Region

avg z-score# best (of 12)

Reference Reg


VOI Method


Combined AV / PiB

White Matter AV / PiB

Cerebellar Gray AV / PiB

Pons AV / PiB

7.20 / 9.94 6.74 / 9.53 5.34 / 7.86 3.94 / 6.64

12 / 9 6 / 6 0 / 3 1 / 0

MR mask

AV / PiB

FDG mask

AV / PiB

Atlas VOIs

AV / PiB

6.90 / 9.40 6.10 / 8.50 4.43 / 7.58

17 / 15 6 / 7 1 / 2

All Measures AV / PiB

5.81 / 8.49

5 / 67

Radiotracer


Partial Volume Correction

• Benefit: Subjects with increased amyloid may likely have more gray matter tissue loss (hence more significant partial volume effects) than subjects with little to no amyloid load. Hence, PVC will enhance the PET signal in amyloid positive subjects.

• Concern: Errors in segmentation of the MR into gray matter, white matter, CSF, and other regions will propagate into estimates of the PV corrections.

• Key Question (trade-off): Does the difference in tissue atrophy between groups or individuals effect the magnitude partial volume effect by enough to offset the increased variance associated with errors the correction?

Partial Volume Correction

• 29 subjects with PiB and AV45 scans: 12 negative; 17 positive

• Gray and white matter segmentation determined by FreeSurfer.

• Masks are smoothed to scanner-specific PET resolution producing gray and white matter fraction images (fg and fw).

• VOIs applied to the co-registered amyloid images are also applied to fg and fw.

• White matter value (WM) based on average of all voxels with >90% contribution from white matter.

• Partial volume-corrected cortical VOIs based on:

VOIPVC = [VOIraw – fw(voi) WM] / fg(voi)


Co-registered AV45 PET

Grey matter mask from segmentation

White matter mask from segmentation

White matter mask smoothed to PET resolution

Grey matter mask smoothed to PET resolution

Determination of White Matter Value

• Take smoothed white matter mask and define white matter VOI as all voxels with >90% contribution from white matter

MR-based Tissue Fractions for Cortical VOIs

Partial Volume Fraction

Lateral Frontal

Medial Frontal

Posterior Cingulate

Superior Parietal


Gray (mean ± sd)Low (n=12) 0.56±0.04 0.55±0.04 0.57±0.04 0.55±0.04 0.60±0.03 0.51±0.04

** High (n=17) 0.52±0.03 0.51±0.04 0.50±0.05 0.50±0.04 0.56±0.04 0.48±0.04

White (mean ± sd)Low (n=12) 0.23±0.05 0.24±0.05 0.21±0.04 0.25±0.05 0.21±0.04 0.29±0.04

High (n=17) 0.25±0.03 0.27±0.05 0.26±0.06 0.29±0.05 0.25±0.04 0.31±0.06

CSF (mean ± sd)Low (n=12) 0.21±0.02 0.21±0.02 0.22±0.02 0.20±0.03 0.19±0.02 0.20±0.02

High (n=17) 0.23±0.04 0.22±0.06 0.24±0.03 0.21±0.05 0.19±0.04 0.21±0.03

** Gray matter VOI tissue fractions average 8% lower in the “high” amyloid group than the “low” amyloid group

Partial Volume Correction Amyloid Measures (Temporal Cortex)

PVC effect on individual Z-score measures (Cortex: ‘Avg. 5’)

Take-home Points and Parting Thoughts

• Reference Regions: While cerebellar gray may in theory be the most suitable reference region, its dependence on scatter correction, plus issues with FOV and noise in lower slices suggest that a combined reference region should be used in amyloid imaging. White matter works well, but needs to be defined on co-registered MR.

• Target Regions: For the Alzheimer-type dementia, the standard regions work well. Region definition and data extraction based on co-registered MR or co-registered FDG is superior to region definition on amyloid images alone.

• Positivity thresholds: Robust identification of the reference region is the most crucial step for quantitative determination of amyloid positivity. For visual interpretation, target region and white matter are most important.– The ability to identify amyloid positive subjects may enhanced by surface maps,

which are better suited to visualizing the binding pattern by removing white matter signal

Take-home Points and Parting Thoughts

• Scanners/software: Reproducible differences in amyloid images are seen across vendors, models, and software, which are most likely due to differences in scatter correction implementation on the different systems– Slightly higher contrast in GE and Siemens PET/CT systems, slightly lower in

Philips and Siemens PET-only ECAT systems

– HRRT with older software, highest contrast of all; newer software, lowest of all

• AV45 vs PiB: The primary concern of AV45 relative to PiB is not higher white matter signal, but the lower cortical signal (~60% of PiB) from amyloid– Visual readers of amyloid images must be cognizant of this fact, but will be aided

by the better noise characteristics of a F-18 labeled radiotracer.

• Partial Volume: PVC may be good for assessing atrophy differences between groups, but in most cases, PVC will not aid determining amyloid positivity.– Without PVC: quantitative accuracy limited by variability in the reference region.

– With PVC: quantitative accuracy limited by PVC errors

Technology

Koeppe ppt