Arna van Engelen1, Torben Schneider2, Hubrecht de Bliek3, Miguel Silva Vieira4, Isma Rafiq4, Tarique Hussain4, Rene Botnar1, and Jordi Alastruey1
1Department of Biomedical Engineering, Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom, 2Philips Healthcare, Guidford, United Kingdom, 3Philips Healthcare, HSDP Clinical Platforms, Best, Netherlands, 4Department of Cardiovascular Imaging, Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom
Synopsis
Accurate 3D length measurements through the
aorta are required for Pulse Wave Velocity (PWV) measurements. We evaluate
automatic centreline tracking, requiring only a start and end point, on three different
types of MR data (balanced-SSFP, contrast-enhanced and black-blood MRI), in 12
elderly subjects and 10 patients post-coarctation repair. Our algorithm uses
vesselness filtering, fast marching and centreline refinement. Length differences
between manual and automatic centrelines
are generally below 1cm, with corresponding PWV
differences well below 0.5m/s. This shows that with minimal user interaction,
accurate PWV measurements can be performed using automatic centreline tracking, on
commonly used types of MR data.Purpose
Aortic stiffness is an
important biomarker for a variety of cardiovascular diseases, and can be assessed
by pulse wave velocity (PWV). PWV can be derived from MRI by computing the
blood flow profile at two locations in the aorta and the distance between those
locations. For accurate distance measurements, the aortic centrelines need to
be extracted from 3D images. Automatic aortic centreline tracking has extensively
been evaluated on CTA data
1, but limited studies exist for MRI
2. The performance of automatic algorithms depends
on the input provided and often needs to be optimised for different MR
contrasts. The aim of this study was to develop an aortic centreline tracking
algorithm that performs accurately on images from the most common cardiac MRI
sequences.
Methods
Data
We included 12
subjects from a twin cohort3 and 10 non-stented patients post-coarctation
repair. All subjects underwent phase-contrast velocity-encoded cine in the
ascending and diaphragmatic aorta (125 reconstructed phases). Anatomical sequences
used for 3D centreline tracking included DIR-TSE black-blood images for the
twins and both 3D balanced SSFP (bSSFP) and 3D contrast-enhanced MRA (CE-MRA) timed
for optimal aortic enhancement, for the coarctation patients (Figure 1). All
images were acquired on a 1.5T Philips Ingenia scanner.
Centreline tracking
This consists of three
steps: 1) vesselness filter, 2) fast marching and 3) centreline refinement. The
potential of the vesselness filter has been demonstrated before4. It
uses the Hessian matrix, composed of local second-order derivatives of the
image, to enhance vessel-like structures. We compared several scale settings
for the Hessian matrix: 4 scales, ranging from 4 to 7mm, and 2 scales being
either 4 and 6 or 6 and 8mm.
The start and end
points for centreline tracking were defined by taking the centre of the aorta
on the first phase of the phase-contrast images. An ellipse was fitted on the
3D data at these points and the centres were used as start and end points, to
account for patient displacement during scanning. Bi-directional fast marching5
was performed from both start and end. Finally, the obtained centrelines were centred
and smoothed by an open active contour6. Intensity of the black-blood images was
inverted before centreline tracking.
Flow waveforms and
PWV
Volumetric flow waveforms
were obtained from phase-contrast MRI at the ascending and diaphragmatic aorta,
by fitting the vessel edge along a number of ray casts from a propagated centre
point on all phases7. The arrival of the pulse wave was determined
by determining the foot of the curve, and transit time was determined by taking
the time difference between the two feet8. PWV was calculated as the
ratio of the centreline length to the transit time.
Evaluation
Manual centrelines
were annotated three times by the same observer in all anatomical scans using
all three imaging planes. Centrelines were resampled to 0.1mm and the manual
centrelines were cropped from the points closest to the end points of the
automatic centrelines. Centrelines were evaluated based on success (remaining
inside the lumen), length, point-based distance to the manual centreline, and
the effect on PWV measurements. For the coarctation patients the difference in
length between bSSFP and CE-MRA was also evaluated.
Results and discussion
Quantitative results
are provided in Tables 1 and 2, and examples of obtained centrelines are shown
in Figures 2 and 3.
For the black-blood
and CE-MRA data, length differences generally stay below 1cm, resulting in PWV
differences well below 0.5m/s, being clinically acceptable. For bSSFP data the
differences are slightly larger, which is mostly attributable to one case where
the tracked centreline followed a wrong path. Computing the Hessian at two
scales (4-6mm) yielded best accuracy for all image types.
Length differences
between bSSFP and CE-MRA can be due to differences in centreline accuracy and
image characteristics, as well as patient displacement (Fig 2). However, these differences
stay within acceptable ranges.
In practice, manual
correction of inaccuracies on the obtained centrelines is feasible, so a
semi-automatic approach is possible and would improve PWV accuracy in such
cases.
Conclusion
This semi-automatic aortic
centreline tracking technique performs well for the three most commonly used
cardiac MRI sequences. The obtained centrelines are suitable for accurate aortic
PWV measurements.
Acknowledgements
This research has been supported by an
EPSRC Technology Strategy Board CR&D
Grant (EP/L505304/1).References
1.
Worz
et al., IEEE Trans Biomed Eng 2010
2.
Babin
et al., Conf Proc IEEE EMBS 2012
3. Moayyeri et al., Int J of
Epidemiology 2013
4. Frangi et al., MICCAI 1998
5. Wink et al., PhD thesis 2004
6.
Lobregt
et al., IEEE Trans Med Imaging 1995
7.
Wink
et al., IEEE Trans Med Imaging 2000
8.
Gaddum
et al., Ann Biomed Eng 2012