Multisite feasibility study of spinal cord gray matter and total cord areas measurements on 2D Phase Sensitive Inversion Recovery images
Nico Papinutto1, Esha Datta1, Alyssa H Zhu1, Julio Carballido-Gamio2, Regina Schlaeger1,3, Sinyeob Ahn4, Kevin Johnson4, Lara Stables5, William A Stern1, Gerhard Laub4, and Roland G Henry1

1Neurology, University of California San Francisco, San Francisco, CA, United States, 2Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States, 3Neurology, University Hospital Basel, University of Basel, Basel, Switzerland, 4Siemens Healthcare USA, San Francisco, CA, United States, 5Neuroscience Imaging Center, University of California San Francisco, San Francisco, CA, United States

Synopsis

The goal of the present work was to test the reliability of the C2-C3 spinal cord gray matter and total cord areas measurements performed using a 2D-PSIR sequence. Nine healthy subjects were scanned twice with repositioning in between the scans (test/retest) on three different 3T scanners with different hardware. On the phase sensitive-reconstructed images, total cord area was measured in a semi-automated way and gray matter area was estimated by using an automatic segmentation method. Evaluations of contrast to noise ratio, intra-scanner and inter-scanner reliability suggest that multicenter studies using a 2D-PSIR sequence are feasible.

Purpose

To evaluate the intra- and inter-scanner reliability of spinal cord (SC) total cord area (TCA) and gray matter (GM) area measurements based on 2D-Phase Sensitive Inversion Recovery (2D-PSIR) imaging at 3T.

2D-PSIR has been recently shown to be very promising for reliable measurements of SC TCA and GM area on healthy controls and multiple sclerosis patients1,2. The goal of the present work was to test the reliability of the SC areas measurements at the C2-C3 disc level on different scanners, each with different hardware, at different sites.

Methods

C2-C3 images of nine healthy controls (age 33±10 [mean±SD]; 5 females/4 males) were acquired at 3 different sites, positioning the 2D-PSIR single slice perpendicularly to the SC on a standard sagittal T2-w image1. In each session each subject was scanned twice with repositioning in between the scans (test/retest).

2D-PSIR sequence parameters were: 0.78x0.78x5 mm3 spatial resolution, TR/TE/TI=4000/3.22/400 ms, flip angle=10°, and 3 averages (acquisition time: 1:50 min, magnitude and phase-sensitive images reconstructed).

The scanners at the different sites were:

SCANNER1: Siemens Skyra 3T, 64-channel head/neck coil, pTX technology

SCANNER2: Siemens Skyra 3T, 20-channel head/neck coil

SCANNER3: Siemens Trio 3T, 12-channel head coil plus 4-element neck coil

Evaluation of contrast to noise ratio (CNR)

The CNR for cerebrospinal fluid/white matter (CSF/ WM) and for WM/GM tissues was calculated on the magnitude-reconstructed C2-C3 images.

CNR between tissue 1 and tissue 2 was defined as CNR12=|S1-S2|/BN, where S1 and S2 were the average signals in two identical 2x2 voxels square ROIs positioned on the tissues and BN (background noise) was the standard deviation of the signal measured in a ROI of 100 mm2 outside the neck, away from imaging artifacts. The GM ROI was positioned on the anterior horn, the WM ROI on the lateral column.

Evaluation of intra-scanner and inter-scanner reliability

TCA was measured in a semi-automated way on all the phase sensitive-reconstructed images (9 subjects x 2 images/site x 3 sites =54) by using the software Jim1,3 (www.xinapse.com).

GM area was estimated by using an automatic segmentation method (described elsewhere) based on a publicly available Python implementation of the morphological geodesic active contour method4 (github.com/pmneila/morphsnakes).

Briefly: This method uses the binary spinal cord masks (obtained with the semi-automatic Jim segmentation) and spinal cord GM masks (manually segmented by a neurologist expert in neuroimaging) of C2-C3 PSIR acquisitions on 20 subjects to create a template.

This template is subsequently used in the automatic segmentation of each PSIR acquisition: First, the cord shape template is registered to each subject’s spinal cord shape mask with affine and non-linear transformations. These transformations are then applied to the spinal cord GM template. This initial result is used in an active contour algorithm to obtain the final GM segmentation.

Intra-scanner (between the test/retest acquisitions) and inter-scanner (pairs of test/retest acquisitions on the 3 scanners) reliabilities for TCA and GM areas were calculated in terms of coefficients of variation (COV = standard deviation/mean of the values).

Results

It was possible to segment the TCA in all the 54 images.

On 52/54 images (~96%) the automatic technique yielded visually correct GM segmentations (Figure 1). One acquisition from SCANNER1 and one from SCANNER3 were excluded from the GM analysis since based on visual inspection the segmentations clearly failed (Figure 2).

Evaluation of CNR

The average CSF/WM CNR for the 9 subjects (mean±SD) was:

SCANNER1: 59.18±18.22

SCANNER2: 41.79±8.92

SCANNER3: 42.26±10.07

The average WM/GM CNR was:

SCANNER1: 20.74±7.99

SCANNER2: 14.96±2.30

SCANNER3: 16.60±3.76

Evaluation of intra-scanner and inter-scanner reliability

For TCA the intra-scanner COV was:

SCANNER1: 0.91%±0.58%

SCANNER2: 1.07%±0.92%

SCANNER3: 0.51%±0.49%

The inter-scanner TCA COV was:

Test: 0.97%±0.74%

Retest: 1.07%±0.50%

Mean value of the two acquisitions: 0.90%±0.38%

For GM area the intra-scanner COV was:

SCANNER1: 2.51%±1.91%

SCANNER2: 2.68%±3.20%

SCANNER3: 2.80%±2.78%

The inter-scanner GM area COV was:

Test: 4.98%±1.84%

Retest: 3.37%±2.26%

Mean value of the two acquisitions: 3.60%±1.80%

Discussion and Conclusion

The C2-C3 2D-PSIR images of 9 healthy controls had very similar CNR on the 3 different scanners. As expected due to the advanced hardware, mean CNR for SCANNER1 was the highest in the group.

Semi-automatic TCA and automatic GM area measurements had a very high intra-scanner and inter-scanner reliability. The comparable values obtained for intra- and inter-scanner measurements suggest that multicenter studies using 2D-PSIR are feasible.

A limitation of this study is that scanners of a single vendor were used. Future studies are necessary to verify the feasibility of multicenter studies with different vendors’ scanners, but preliminary data (not shown) suggests the quality of data obtainable with the 2D-PSIR protocol is not vendor specific.

Acknowledgements

No acknowledgement found.

References

1. Papinutto N, Schlaeger R, Panara V, Caverzasi E, Ahn S, Johnson KJ, Zhu AH, Stern WA, Laub G, Hauser SL, Henry RG. 2D phase-sensitive inversion recovery imaging to measure in vivo spinal cord gray and white matter areas in clinically feasible acquisition times. J Magn Reson Imaging. 2015 Sep;42(3):698-708.

2. Schlaeger R, Papinutto N, Panara V, Bevan C, Lobach IV, Bucci M, Caverzasi E, Gelfand JM, Green AJ, Jordan KM, Stern WA, von Büdingen HC, Waubant E, Zhu AH, Goodin DS, Cree BA, Hauser SL, Henry RG. Spinal cord gray matter atrophy correlates with multiple sclerosis disability. Ann Neurol. 2014 Oct;76(4):568-80.

3. Horsfield MA, Sala S, Neema M, Absinta M, Bakshi A, Sormani MP, Rocca MA, Bakshi R, Filippi M. Rapid semi-automatic segmentation of the spinal cord from magnetic resonance images: application in multiple sclerosis. Neuroimage. 2010 Apr 1;50(2):446-55.

4. Márquez-Neila P, Baumela L, Alvarez L. A morphological approach to curvature-based evolution of curves and surfaces. IEEE Trans Pattern Anal Mach Intell. 2014 Jan;36(1):2-17.

Figures

Figure 1: Example of gray matter automatic segmentation results on the 6 acquisitions performed on the same subject.

Figure 2: The two cases (~4% of total) in which the automatic gray matter segmentation failed.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
4409