IMPACT OF PARALLEL IMAGING ON VOLUMETRIC BRAIN MRI MEASUREMENTS IN ADNI-GO AND ADNI-2

S. Gouttard , J. Schaerer , H.J. Yu , A. Istace , F. Roche , C. Pachai , L. Bracoud and the Alzheimer's Disease Neuroimaging Initiative
1 1 1 1 1 1 1 1

BioClinica, Lyon, France & Newtown, PA, USA - email: Sylvain.Gouttard@bioclinica.com

INTRODUCTION
Parallel MR imaging has become increasingly available in recent years. Its use allows decreasing overall scan time and as a consequence the risks of motion artifacts which can dramatically impair the accuracy of the volumetric measurements performed out of volumetric MRI (3D T1-weighted) data. But the impact of parallel imaging/acceleration on volumetric assessments requires further investigation, as it may balance potential gains in accuracy. The objective of this study was to assess the impact of acceleration on classical volumetric parameters (volumes of hippocampus, whole brain and lateral ventricles) in the context of multi-center clinical trials.

True positive rate

Statistical analysis Pearsons correlation was estimated between accelerated and non-accelerated results for normalized WBV, LVV and HCV. A Receiver Operating Characteristic (ROC) analysis aiming at separating between NC and AD subjects was performed for the whole set of accelerated and non-accelerated data, as well as for 1000 random combinations of both sorts, in order to simulate the context of a clinical trial combining MRI scans from various sites with or without parallel-imaging capabilities.

ROC analysis led to similar areas under the curve (AUC) when attempting to separate NC and AD subjects using accelerated data, non-accelerated data or a random combination of both sorts (minimum value of 1000 random combinations is reported) (see Figure 3).
1.00

Acc.
0.50

Non-acc. 0.82 0.78 0.91

Random 0.81 0.78 0.90

Whole Brain
WBV - Accelerated WBV - Non-accelerated LVV LVV - Accelerated - Non-accelerated

0.82 0.79 0.92

RESULTS METHODS
Population The population used for this study was composed of ADNI-GO and ADNI-2 subjects. 3DT1 sequences used a sagittal 3D Fast SPGR (General Electric), 3D TFE (Philips) or MP-RAGE (Siemens) pulse sequences, performed on 3T scanners. Accelerated data used a resolution of 1.1 mm x 1.1 mm x 1.2 mm. Acceleration factor was set to 1.75 (GE), 1.8 (Philips) and 2 (Siemens) depending on the manufacturer. Non-accelerated data used a resolution of 1.0 mm x 1.0 mm x 1.2 mm. 291 subjects from ADNI-GO and ADNI-2, scanned with both accelerated and nonaccelerated 3DT1 sequences, were selected for this study. The population included: 82 Normal Controls (NC) 102 Early Mild Cognitive Impairment (EMCI) 59 Late MCI (LMCI) 48 Alzheimers Disease (AD) Image analysis Accelerated and non-accelerated sequences were processed separately, using fully automated methods. The volumes of whole brain (WBV), lateral ventricles (LVV) and hippocampus (HCV) were automatically quantified using previously reported atlas-based segmentation algorithms [1,2]. Intracranial cavity volume (ICV) was automatically quantified as well using an atlas segmentation algorithm [3] in order to normalize the above-mentioned measurements. Correlation between accelerated and non-accelerated results Accelerated and non-accelerated data were strongly and significantly correlated for each structure (r > 0.98, p < 0.001, see Figure 1).
Whole Brain
WBV (%ICV) - Non-accelerated
0.80

Lateral Ventricles Hippocampus

HCV - Accelerated HCV - Non-accelerated

0.00 0.00 0.50 1.00

False positive rate

Lateral Ventricles
0.007

Hippocampus
HCV (%ICV) - Non-accelerated

Figure 3 - ROC curve analysis (left) and the associated AUC values (right) for each normalized measurement

LVV (%ICV) - Non-accelerated

0.09

r = 0.991 p < 0.001

0.70

r = 0.998 p < 0.001

0.03

r = 0.988 p < 0.001

CONCLUSIONS

0.004

0.70

WBV (% ICV) - Accelerated

0.80

0.03

LVV (% ICV) - Accelerated

0.09

0.004

HCV (% ICV) - Accelerated

0.008

Figure 1 - Correlation between accelerated and non-accelerated results

Group separation Normalized volumetric measurements showed similar distributions across groups and structures (see Figure 2).
Whole Brain
0.80 0.09

Normalized volumetric MRI measurements (WBV, LVV and HCV) obtained using accelerated and non-accelerated data were strongly and significantly correlated and exhibited a similar ability to separate NC and AD subjects. This supports the idea that the use of parallel imaging in clinical trial MR protocols with a volumetric component should be highly encouraged for faster acquisition and decreased occurrence of motion artifacts. The impact on subsequent atrophy measurements will have to be assessed in the future.

Lateral Ventricles
0.007

Hippocampus

REFERENCES
[1] Belaroussi et al., Labeling of brain MRI images using atlas propagation and classificationbased nearest-neighbor transform, Poster session, AAN 2011 [2] Belaroussi et al., Multi-Atlas hippocampus segmentation refined with intensity-based tissue classification, Poster session, AAIC 2012 [3] J. Schaerer et al., Accurate intra-cranial cavity delineation and volume measurement in the ADNI-1 MCI population using multi-atlas segmentation, Poster session, AAIC 2012 bioclinica.com/aaic2013-gouttard-poster
HCV (% ICV)
0.004

WBV (% ICV)

0.70

Accelerated

Non-accelerated

LVV (% ICV)
0.03

Accelerated

Non-accelerated

Accelerated

Non-accelerated

NC

EMCI

LMCI

AD

Figure 2 - Volumes of each structure of interest for accelerated and non-accelerated scans

Alzheimers Association International Conference (AAIC) July 13 18, 2013 Boston, USA