Yuichi Suzuki1, Akira Kunimatsu1, Kouhei Kamiya1, Masaki Katsura1, Harushi Mori1, Katsuya Maruyama2, Thorsten Feiweier3, Kenji Ino1, Yasushi Watanabe1, Jiro Sato1, Keiichi Yano1, and Kuni Ohtomo1
1The Department of Radiology, The University of Tokyo Hospital, Bunkyo-ku, Japan, 2Siemens Japan K.K., Shinagawa-ku, Japan, 3Siemens AG, Erlangen, Germany
Synopsis
We quantitatively evaluated the quality of
tractography images captured using the multi-band EPI (MBEPI) compared with those obtained without using
MBEPI. We also demonstrated the potential weakness of classic Dice similarity
coefficients (DSCs) and introduced a structural similarity (SSIM) index and map as a new method for evaluating the quality of tractography images. A numerical evaluation was enabled by the SSIM index
and that the SSIM map was advantageous in that it allows visual confirmation of the
structural similarity ratio; in contrast, the DSCs only offered a numerical
evaluation.Purpose
The multi-band
echo planar imaging (MBEPI) technique 1,2 can reduce scanning time; however, the quality
of the images produced is suboptimal because of the effects of incomplete slice separation 3. In this study, we quantitatively evaluated the
quality of tractography images captured using MBEPI compared with those
obtained without using MBEPI. We also demonstrated the potential weakness of classic
Dice similarity coefficients (DSCs) and introduced a structural similarity
(SSIM) index and map 4 as a new method for evaluating the quality of
tractography images.
Materials and Methods
Twelve
men participated in this study (average age, 28.25 ± 3.62
years). Diffusion MRI (dMRI) data were acquired with a SIEMENS MAGNETOM Avanto
1.5T B17 using a single-shot spin echo EPI sequence and following parameters: b-value,
3000 s/mm2; field of view, 24.5 × 24.5 cm2; acquisition
matrix, 98 × 98; slice thickness, 2.8 mm; axial sections, 50; GRAPPA factor, 2; and average, 1.
The following three methods were compared.
Non-MBEPI method (gold standard): MPG, 64
axes and acquisition time, 518 s.
MBEPI method 1 (shortened scanning time): MPG,
64 axes; MB factor (MBf), 2; and acquisition time, 327 s.
MBEPI method 2 (an increased number of MPG
directions): MPG, 104 axes; MBf, 2; and acquisition time, 507 s.
The MBEPI
sequence version used was Release R011a for VB17A 5.
After the
distortion correction of dMRI data using eddy_correct in FSL, and Q-ball
imaging (QBI) analysis and tractography were performed using Diffusion Toolkit
0.6 and TrackVis 0.5.1. The tractography of the bilateral pyramidal tract was
reconstructed using the standard method; by placing the seed and target in the
cerebral peduncle and motor cortex, respectively 6. The quality of the
tractography images obtained using MBEPI methods 1 and 2 was evaluated using
the non-MB method as the gold standard. The DSCs and SSIM index were used as
quantitative measures of coincidence with the non-MBEPI method. A visual
assessment (VA), a method similar to the gold standard was performed by a neuroradiologist
and radiologic technologist with 10 years’ experience in neurosurgery MRI. The
SSIM index is generally used to measure the similarity between two images or
videos and has been widely applied in the evaluation of image compression 7,
8. The SSIM index can be considered a measure of the quality of one of
the images under comparison, provided that the other image is of perfect
quality.
We performed McNemar's test using the VA and the DSCs
or the SSIM index and calculated the coincidence rate (accuracy) between the VA
and the DSCs or the SSIM index with SPSS v.20. Both the DSCs and SSIM index were
used to generate an SSIM map by voxel-wise calculation using MATLAB codes.
Results
The DSCs for coincidence with the non-MBEPI
method of the right/left pyramidal tracts obtained using MBEPI methods 1 and 2
were 0.614/0.618 and 0.609/0.647, respectively, of the DSC obtained using the non-MB
method.
The
SSIM indices of MBEPI methods 1 and 2 were 0.936/0.942 and 0.935/0.947,
respectively (Fig.1). McNemar’s test identified no significant differences between the VA
and DSC (p = 0.143) or the VA and SSIM index (p = 0.267). The accuracies were
0.646 and 0.729 for the VA and DSC, and the VA and SSIM index, respectively (Fig.2). It
was easy to visualize the similarity ratio using the SSIM map (Fig.3).
Discussion and Conclusion
The results of McNemar's test and coincidence rates
showed that the SSIM index is more similar to the VA than the DSC. This result
supports Wang’s suggestion
1 that the SSIM calculates a value almost
equivalent to that selected by subjective judgment. We believe that a numerical evaluation is enabled
by the SSIM index and that the SSIM map is advantageous in that it allows
visual confirmation of the structural similarity ratio; in contrast, the DSC only
offers a numerical evaluation. Therefore, this initial study demonstrates the utility
of the SSIM index and map for the evaluation of the quality of tractography images;
moreover, the success of the SSIM index and map may preface the development of
new methods for the evaluation of tractography. Regarding MBEPI methods 1 and 2, there were no great
differences between these methods of evaluation. However, it’s difficult to declare that
the result with 64 axes MPG set is the gold standard for QBI tractography.
Therefore, examinations with other conditions are required in future works.
Acknowledgements
I am deeply
grateful to Yuji Komaki of
Central Institute for Experimental Animals.
This study was supported by Grants-in-Aid for
Scientific Research (26460721).
References
1. Larkman DJ, deSouza NM, Bydder M, et al. An investigation into the use of sensitivity-encoded techniques to increase temporal resolution in dynamic contrast-enhanced breast imaging. JMRI. 2010; 13(2):313-317.
2. Moeller S, Yacoub E, Olman CA, et al. Multiband multislice
GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel
imaging with application to high spatial and temporal whole-brain fMRI. Magn Reson Med. 2010; 63(5):1144-1153.
3. Xu J, Moeller S, Auerbach EJ, et al. Evaluation of slice accelerations using multiband echo planar imaging at 3 T. Neuroimage. 2013; 83:991-1001.
4. Wang Z, Bovik AC, Sheikh HR, et al. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on
Image Processing, 2004 Apr; 13(4):600-612.
5. https://www.cmrr.umn.edu/multiband/
6. Wakana S, Caprihan A, Panzenboeck MM, et al. Reproducibility of quantitative tractography methods applied to cerebral white matter. Neuroimage. 2007 July
1; 36(3):630–644.
7. Coulombe S, Pigeon S. Low-complexity transcoding of JPEG images with near-optimal quality using a predictive quality factor and scaling parameters. IEEE Trans Image
Process. 2010 Mar; 19(3):712-21.
8. Voruganti AK, Mayoral R, Vazquez A, et al. A modular video streaming method for surgical assistance in operating room networks. Int J Comput Assist
Radiol Surg. 2010 Sep; 5(5):489-99.