Quality Control measures for Constrained Spherical Deconvolution MR diffusion tractography in clinical use.
Donald W McRobbie1,2 and Marc Agzarian1

1Medical Imaging, Flinders Medical Centre, Adelaide, Australia, 2Surgery, Imperial College, London, United Kingdom


Quality Control (QC) methods for clinical MR tractography using whole-brain Constrained Spherical Deconvolution (CSD) in individual patients are used to assess the quality of the acquired data. Clinical scoring of the resulting tractograms demonstrates robust depiction of anatomically realistic tracts over a range of MR scanners, acquisitions, and with varying raw image quality. Whole brain CSD shows potential for clinical use subject to suitable QC measures.


One of the barriers to the clinical utilization of MR diffusion tractography lies in the uncertainty regarding how the tractography algorithm handles white matter (WM) fibres crossing within a voxel. Recently Constrained Spherical Deconvolution (CSD) has provided a means to resolve the directionality issues inherent in conventional Diffusion Tensor Imaging (DTI) and has demonstrated anatomically realistic tracts1,2. A further problem in clinical use is the requirement to manually seed the starting and/or target point for tracking, adding a degree of operator dependence. CSD can remove this potential source of error using whole-brain tracking. Despite these improvements, there is still uncertainty about results for the individual patient: both from false positives (“tracks” that are artefactual) and false negatives (non-depiction erroneously concluded to equal non-existence). To improve confidence in single patient CSD, we devised a simple quality control (QC) procedure using the patient’s images.


Three female and six male patients (age: mean=21.7, SD=13.3, range 3-43 yr) underwent multi-directional diffusion imaging (15, 30, 33 or 64 directions) on a 1.5T (Philips Ingenia, Siemens Aera) or 3T (Siemens Trio) MR with b-values of 800,1000 or 3000 s mm-2. Pixel size was 1.8mm2 , slice width 2.5 or 5 mm, parallel imaging reduction factor 2, and TE in the range 82-114 ms.

CSD tractography was performed in MRtrix3 using a whole-brain mask, seed and target. The single fibre (SF) response function was computed for a harmonic order of 8 for the 64 direction images, 6 for 30 and 33 directions, and 4 for 15 directions. A fractional anisotropy (FA) threshold of 0.7 determined the region-of-interest (ROI) for calculation of the SF response function. 40,000 whole-brain tracks were plotted using probabilistic streaming.

Clinical evaluation (0-10) required demonstration of:

1. Left, right corticospinal tracts as two discrete structures in the anterior medulla and pons (CS).

2. Left, right medial lemnisci as two discrete structures in the posterior medulla and pons (ML).

3. Transverse pontine fibres separating the corticospinal tracts from the medial lemnisci (PF).

4. Decussation of the superior cerebellar peduncles in the midbrain (CP).

5. Left, right optic tracts extending from the optic chiasm to the left and right lateral geniculate bodies (OT).

6. Left, right optic radiations extending from the left and right lateral geniculate bodies to the left/right primary visual cortex including Meyer loops (OR).

7. Anterior commissure (AC).

8. Left, right superior longitudinal fasciculi (LF).

9. Corpus callosum (CC).

10. Subcortical U-fibres bilaterally (UF).

Quantitative QC utilised (a) whole-brain SNR from the b=0 images, (b) Contrast and CNR of the SF-ROI for the average diffusion images, (c) FA histogram (whole-brain), (d) percentile of FA threshold wrt the FA histogram, (e) median, mean, standard deviation of FA in a ROI positioned within cerebrospinal fluid (CSF).


Representative slices for patients scored as 9 and 10 are shown in figure 1. Patient 2 exhibited excessive motion and scored poorly for anatomy (score=3.5). The 15 direction CSD was also poor (score=5). 6/7 remaining patients had scores of 9 or above, with 1/7 scoring 8. Figure 2 summarizes the demonstration of anatomical features in the study. Depiction of the Meyers loop in the optic radiation was the least well demonstrated feature.

SNR, contrast and CNR are shown in figure 3. Contrast (but not SNR or CNR) was improved for b=3000. Patient 2's images (the worst tractogram) had very low contrast and CNR. Figure 4 shows the whole-brain FA histograms with the SF threshold indicated. The apparent FA of an isotropic medium (CSF) is shown in figure 5, with the upper percentile FA for the SF threshold.


The measured FA for CSF gives an indication of the ‘directional noise’ in the image. The median CSF-FA value decreased with increased b-value, but did not necessarily result in better anatomical scores. Low angular resolution and excessive movement resulted in reduced anatomical scores. Similarly image SNR, contrast and CNR did not strongly correlate with the anatomical ranking, although the lowest contrast and CNR did result in the worst quality tractogram. Although the only perfect score was obtained for a 3T, b=3000, n=64 acquisition, diagnostically acceptable tractograms were achievable for acquisitions made at 1.5T with lower b-values. Previously the quality of CSD tractograms has been invested with phantoms and theoretically4.


Despite a significant range of acquisition conditions, ages, and clinical details resulting in a spread of objective QC metrics, whole-brain CSD demonstrated WM tract anatomy consistently. 7/9 tractograms yielded acceptable clinical information. Whilst a larger study is needed, these results show the potential for clinical CSD-tractography to be used on an individual patient basis.


The authors are grateful to SA Health for support, and to Angela Walls for the MR acquisitions.


1. Tournier JD, Calamante F, Gadian DG, Connelly A. Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution. Neuroimage (2004) 23:1176–1185.

2. Tournier JD, Mori S, Leemans A. Diffusion Tensor Imaging and Beyond. Magnetic Resonance in Medicine (2011) 65:1532–1556.

3. Tournier JD, Calamante F, Connelly A. MRtrix: diffusion tractography in crossing fire regions. Int U IMag Syst Technol (2012) 22:53-66.

4. Tournier JD, Yeh CH, Calamante F et al. Resolving crossing fibres using constrained spherical deconvolution: Validation using diffusion-weighted imaging phantom data. NeuroImage (2008) 42:617–625.


Figure 1.Tractogram for patient 9. 3T, b=3000, n=64 (left). Tractogram for patient 4. 1.5T, b=800, n=33 (right).

Figure 2. Depiction of WM tracts. CS-corticospinal, LM-medial lemnisci, PF-pontine fibres, CP-cerebellar peduncles, OT-optic tract, OR-optic radiation, AC-anterior commissure, LF-longitudinal fasciculi, CC-corpus callosum, UF-subcortical U-fibres.

Figure 3. SNR -whole brain from the b=0 image, Contrast and CNR - from the single fibre ROI wrt the average diffusion image. It was not possible to calculate SNR or CNR for the images that utilized SENSE.

Figure 4. Whole brain histograms of FA values, showing the threshold for the single fibre ROI.

Figure 5. Percentile of FA values comprising the single fibre ROI (left scale). Median and standard deviation of FA values from the CSF ROI (right scale).

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