ELSEVIER Basic Science Researchthales.iacm.forth.gr/~yannisp/Journal/AVSG2018.pdf · thrombus deposition was mainly anteriorly accumulated, and there was a posterior thrombus ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
ARTICLE IN PRESS
ELSEVIER
Basic Science Research
Correlation of Intraluminal Thrombus Deposition, Biomechanics, and Hemodynamics with Surface Growth and Rupture in Abdominal Aortic Aneurysm-Application in a Clinical Paradigm
and Yannis Papaharilaou, 1 Heraklion, Crete, Greece
Background: The natural history of abdominal aortic aneurysm (AAA) can be investigated through longitudinal evaluation of localized aneurysm characteristics exploiting clinical images. The major challenge is to identify corresponding regions between follow-ups. We have recently developed an algorithm (VascForm) based on nonrigid registration that can obtain surface correspondence and quantify surface growth distribution. Methods: A ruptured AAA with an initial computed tomography scan 2 years ago was studied. Following 3-dimensional reconstruction of outer wall and luminal surfaces, the wall/thrombus thickness was obtained. Wall stress distribution was computed with finite element analysis, and computational fluid dynamics were performed. VascForm was applied and allowed for the ruptured wall site to be traced back to the initial wall surface and be correlated with local initial intraluminal thrombus thickness, wall stress, and hemodynamic parameters. It also allowed for the quantification of wall surface growth based on the surface element growth. Results: Rupture occurred at the posterolateral side. Initial wall surface growth was in most regions 40%. However, a large section of the posterior wall presented 110% growth. Initial thrombus deposition was mainly anteriorly accumulated, and there was a posterior thrombusfree isle. Peak wall stress (initial and follow-up) occurred at AAA neck. Nonrigid registration revealed that rupture originated from the vicinity of the initial thrombus-free isle. Furthermore, rupture occurred at the wall region with the largest growth (110%). No clear correlation between hemodynamics and rupture site could be identified. Conclusions: High local surface growth correlates with rupture site and could therefore potentially become a marker of rupture risk. The ongoing application of this methodology to a large cohort of AAA patients will focus on identifying characteristic features of AAA regions that present high surface growth in follow-up evaluations, to assist in improved rupture risk estimation.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
1 Institute of Applied and Computational Mathematics, Foundation
q2 for Research and Technology-He/las, Heraklion, Crete, Greece.
Correspondence to: Yannis Papaharilaou, Nikolaou Plastira 100, VassilikaVouton, Heraklion, Crete GR 700 13, Greece; E-mail: [email protected]
2Vascular Surgery Unit, Department of CardioThoracic and Vascular Surgery, University of Crete Medical School, Heraklion, Crete, Greece. Manuscript received: May 23, 2017; manuscript accepted: August 1,
Contrast-enhanced high-resolution spiral CT angiography images were processed using ITK-SNAP,as previously described.8• 15 Briefly, from the firstCT scan, a 3-dimensional AAA model of the abdominal aorta was reconstructed from the stack of con -tours of the outer wall and lumen surfaces thatwere manually segmented. The reconstructed 3Dsurfaces were processed using the VascularModeling Tool Kit (VMTK). 16 The distance betweenthe outer wall surface and the luminal surface provided the thrombus-wall thickness. Nonrigid Registration
The VascForm algorithm is written in Matlab andis based on nonrigid point cloud registration andspecifically on iterative closest point algorithm.It is adapted to the needs of aneurysm follow-upstudies and has been recently validated. 10 Inshort, it consists of 2 phases. During the firstphase, surfaces are prealigned applying principalcomponent analysis, and a general deformationof the source surface is performed to best matchthe Target using the Procrustes algorithm. Duringthe second phase, surfaces are finely matchedthrough a nonrigid local deformation model.Each point in the source surface is placed closer(with respect to Euclidean distance) to the targetsurface in the following manner: for every pointp on the source surface, the K-nearest neighbors(N
p) are considered, and each neighbor influences
the displacement of p by an amount which is determined via a Gaussian radial basis centered Q3 on p acting as weight function. The input to VaseForm was the initial and follow-up wall surfaces inSTL format. After registration, surface growthdistribution is the element surface area growthdistribution (triangular elements) and the localsurface growth the element area growth. The relative surface growth was computed as the surfaceelement growth between the 2 follow-ups dividedby the initial element surface. The main uncertainty/error of the method depends on the initial surface distance. Specifically, ithas been found to linearly increase with the initialdistance between the surfaces (after pre-alignment).10 The distance index (DI; mm), is used as aglobal metric for the distance between the surfaces:
DI= µDistance+ <rmstance, (1)
where µDistance and <rmstance the mean and standarddeviation of the distance between source surface
Surface growth and rupture in aneurysm 3
al
nodes and target surface. Based on the DI, the error Q4 index graphin (mm2
Metaxa ) can et al. be 10 The obtained error from index the characterrelated
izes face the growth± uncertainty error index), window and of when surface divided growth by (surthe tiveiniti surface element area, it characterizes the relasurface growth uncertainty window (relativesurface growth ± error index/initial surface elementarea). To identify the corresponding regions betweeninitial and follow-up surfaces, it was first necessaryto mark the rupture site on the follow-up wall surface. For this purpose, on 2 consecutive slices ofCT scan where the rupture of the wall was visible,a small mark outside the aorta in front of the rupturewas segmented and saved in the same file of thereconstructed wall surface. Next, VMTK was usedto allow the drawing of the aortic wall region thatwas colocalized with the rupture mark. Thisprovided a surface that had the value of "l" at thenodes of the rupture site and the value of "O" everywhere else. To identify which region of the initial wall surfacehad ruptured, the nodes of rupture site ("l") weretraced back to the initial surface using VascForm. Wall Stress Estimation
Using the outer wall and lumen surfaces and assuminga uniform thickness of 2 mm, the AAA model for finiteelement analysis (FEA) was reconstructed includingboth AAA wall and ILT, as previously described. 17
The 3D mesh was generated using ICEM CFD,v.12.O.1 (ANSYS Inc., Berkeley, CA), and consistedof linear tetrahedral elements with an average meshdensity of 2.4 elements/mm3
, based on previousmesh sensitivity tests. Workbench (ANSYS Inc.) wasused for FEA and estimation of wall stress distributionat peak systole. Hyperelastic material models asdescribed by Raga van and Vorp for wall and VandeGeest for ILT were adopted.3,
4 A uniform 120 mm Hgsystolic pressure wall loading was applied to the lumenboundary, and the result in maximum principle stressdistribution was evaluated. Computational Flow Dynamics
Flow extensions were added to the luminal surfacesof the first and follow-up scans, and a pure hexahedral mesh was constructed using ANSA (BETA CAESystems S.A., Thessaloniki, Greece). The shear thinning property of blood was accounted for, byemploying the Herschel-Bulkley model, µ(,y)= (�)-[1 -exp(-m,y)]+K'Yn-1
Fig. 2. Surface growth distribution of the initial wall surface co-mapped with the follow-up surface (semi-transparent). Most of the surface grew by 40%. A large part of the posterior region though grew by 110 ± 11.8%.
used to obtain correspondence between initial and
follow-up wall surface meshes. It was found that
most of the initial wall surface (mesh element
area) had grown by 40% (Fig. 2). However, there was a large part of the posterior wall that had grown
by ll0%. The error/uncertainty of surface growth estimation was 11.8% (distance index 5.82 mm,
error index of 0.8 mm2, and mean initial element
area 6.8 mm2).
Projection of rupture site on initial wall surf ace.
Since the posterior wall had been displaced far from its initial position (as seen in Fig. 2), locating
the original location of rupture at the initial wall surface was not straightforward. Therefore, Vase
Form was used to trace the rupture site back to the
initial wall surface and search for potential colocalization with aneurysm characteristics. As seen in
Figure 3, the location of rupture was traced back
to its "region of origin" on the initial wall surface.
Thrombus and Wall Stress Distribution
To identify potential triggers for wall rupture,
thrombus and wall stress, evaluated at the initial
and follow-up AAA geometries, were investigated. Initial thrombus deposition was not uniformly
distributed in the sac but was mostly accumulated at the anterior region (Fig. 4A). At the posterior
side, there was an isle where lumen and outer wall
had a distance of 2.2 mm (Fig. 4B). If it is assumed
Surface growth and rupture in aneurysm 5
Final wall surface Initial wall surface
VascForm
...
/SZ.
Fig. 3. Projection of rupture site to the initial wall surface.
that the aortic wall has a thickness of 2 mm, this
isle was thrombus-free and could have been the origin of wall rupture, since, after a manual registra
tion of follow-up to the initial surface (Fig. 4C), there Q5
seemed to be a visual proximity between rupture site and this initial thrombus-free isle. In the follow-up
examination, wall/thrombus thickness was reduced
at the anterior side (Fig. 4D) and not substantially changed at the posterior side (Fig. 4E). Overlay of
the rupture site to the thrombus distribution at the
final examination (Fig. 4F) did not show any
remarkable characteristic of thrombus thickness
that could be associated with rupture. Evaluation of wall stress distribution revealed
that its peak value (0.44 M Pa) occurred distant
from the region of rupture and specifically at the anterior and posterior AAA neck (Fig. 5). At the
follow-up examination, the distribution of wall stress was similar to the initial PWS occurred at
the neck region but with its magnitude reduced to
0.29 M Pa. Robust nonrigid registration revealed that
rupture originated from a region where the wall/
ILT thickness was 3.8 mm (Fig. 6A). Furthermore, rupture occurred at the wall region that grew the
most (Fig. 6B) and specifically ll0%. Having established a correspondence between
initial and follow-up surfaces, it was possible to
also examine the local change in ILT thickness. It was interesting to note that Wall/ILT thickness
increased at the neck region by 5 mm but decreased at the anterior region by 5 mm.
Hemodynamics
Hemodynamic simulation revealed a large spatial
variation of the investigated TAWSS, OSI, and RRT
AVSG3530_proof ■ 25 September 2017 ■ 5/10 ■ ce
496
497 498
499
500
501
502
503
504
505 506
507
508
509
510 5ll
512
513 514
515 516
517
518 519
520
521 522
523 524
525
526 527
528
529 530
531 532
533
534 535
536
537 538
539 540
541
542 543
544
545 546
547 548
549
550
6 Metaxa et al.
A
ANTERIOR POSTERIOR
B
Annals of Vascular Surgery
C
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
Wall/ILT thickness (mm)
10.5-10
I-
z
0.393-
0 a..
(.) -st
.c
ca
·;::
593
594
/Sl.
..X.--,,Y
D E
0...
$
.....J
0 LL
Fig. 4. Wall/thrombus thickness distribution at the
initial AAA evaluation (A, B, and C) and final AAA eval
uation (D, E, and F) at the anterior (A, D) and posterior
(B, E) sides. (A, B) At the initial evaluation, thrombus
was mostly accumulated at the anterior side, whereas
at the posterior side, there was a small isle without
thrombus deposition. (C) Overlay of rupture region
( pointed red disc shape) at final examination (outer semi-
595 (Fig. 8). Interestingly, the thrombus-free isle at the
596 posterior side coincided with low TA WSS and subse-
597 quently with high RRT. The anterior region that was
598 characterized by a substantial decrease in ILT thick-
599 ness presented a marked change in hemodynamics
600 between initial and final examination. Specifically,
601 although at the initial examination, the anterior re-
602 gion experienced low TAWSS, high OSI, and high
603 RRT, as computed to other wall regions, at the final
604 examination, OSI had decreased from 0.49 to 0.1
605 and RRT from 800 (N/m2)-
1 to 20 (N/m2)-
1.
0-
8
transparent surface) on thrombus thickness distribution
at initial AAA surface (inner surface) suggested that
thrombus-free region could potentially coincide with
the origin of rupture. (D, E) At the final evaluation,
thrombus thickness had been increased at the posterior
side but interestingly it had been decreased at the ante
rior side. (F) Rupture location marked on the final wall
surface with ILT distribution data.
DISCUSSION
In the last 20 years, many research efforts have
been performed to advance patient-specific rupture
risk assessment. However, since AAA growth is a
multifactorial process, many identified risk markers
show a potential for clinical use.5'
6'
13'
22 But how
could such a large body of knowledge reach clinical
practice? The filtering, prioritization, and integra
tion of suggested rupture risk markers through sta -
tistical modeling could potentially provide a robust
Fig. 6. Overlay of the rupture's origin (black circle) at the first scan with (A) the initial wall/thrombus thickness distri- 755bution and (B) the surface growth distribution. 756
new methodology for clinical rupture risk assessment. 23 In the meantime, the growth rate of AAA, which is 1 of the 2 current basic risk indicators, could be substantially improved by exploiting advances in medical image processing. In addition, nonrigid registration techniques in combination with longitudinal AAA images can shed light in a vast amount of so far unexplored data that will improve our understanding of AAA natural history. In the present study, the investigation of a ruptured AAA case with a previous 2-year-old CT scan available demonstrates the potential of
generating valuable information through exploitation of nonrigid surface registration, to quantify local surface growth.
The most significant finding of this study is that rupture occurred at a region of high surface growth. This is in line with the current belief that growth in general is a sign of high activity and high rupture risk. This finding alone shows the potentially important impact that local surface growth information could have on clinical rupture risk assessment. Rupture also occurred at the posterolateral side, a common site of rupture. Although most AAAs
1. Darling RC, Messina CR, Brewster DC, et al. Autopsy studyof unoperated abdominal aortic aneurysms. The case forearly resection. Circulation l 977;56:161-4.
2. Martufi G, Lindquist Liljeqvist M, Sakalihasan N, et al. Localdiameter, wall stress, and thrombus thickness influence thelocal growth of abdominal aortic aneurysms. J EndovascTher 2016;23:957-66.
3. Raghavan ML, Vorp DA. Toward a biomechanical tool toevaluate rupture potential of abdominal aortic aneurysm:identification of a finite strain constitutive model and evaluation of its applicability. J Biomech 2000;33:475-82.
4. VandeGeest JP, Sacks MS, Vorp DA. A planar biaxial constitutive relation for the luminal layer of intra-luminalthrombus in abdominal aortic aneurysms. J Biomech2006;39:2347-54.
5. Lee K, Zhu J, Shum J, et al. Surface curvature as a classifierof abdominal aortic aneurysms: a comparative analysis. AnnBiomed Eng 2013;41:562-76.
6. Golledge J, Tsao PS, Dalman RL, et al. Circulating markers ofabdominal aortic aneurysm presence and progression. Circulation 2008; 118:2382-92.
7. Zambrano BA, Gharahi H, Lim C, et al. Association of intraluminal thrombus, hemodynamic forces, and abdominalaortic aneurysm expansion using longitudinal CT images.Ann Biomed Eng 2016;44:1502-14.
8. Kontopodis N, Metaxa E, Pagonidis K, et al. Deformationand distensibility distribution along the abdominal aorta in the presence of aneurysmal dilatation. J Cardiovasc Surg (Torino) 2017;58:72-9.
9. Metaxa E, Kontopodis N, Vavourakis V, et al. The influenceof intraluminal thrombus on noninvasive abdominal aorticaneurysm wall distensibility measurement. Med Biol EngComput 2015;53:299-308.
10. Metaxa E, Iordanov I, Maravelakis E, et al. A novel approachfor local abdominal aortic aneurysm growth quantification.Med Biol Eng Comput 2017;55:1277-86.
11. Tzirakis K, Kamarianakis Y, Metaxa E, et al. A robustapproach for exploring hemodynamics and thrombusgrowth associations in abdominal aortic aneurysms. MedBiol Eng Comput 2017;55:1493-506.
12. Stenbaek J, Kalin B, Swedenborg J. Growth of thrombusmay be a better predictor of rupture than diameter in
13. Metaxa E, Kontopodis N, Tzirakis K, et al. Effect of intraluminal thrombus asymmetrical deposition on abdominalaortic aneurysm growth rate. J Endovasc Ther 2015;22:406-12.
14. Kontopodis N, Metaxa E, Papaharilaou Y, et al. Changes ingeometric configuration and biomechanical parameters ofa rapidly growing abdominal aortic aneurysm may provideinsight in aneurysms natural history and rupture risk. TheorBiol Med Model 2013;10:67.
15. Yushkevich PA, Piven J, Hazlett HC, et al. User-guided 3Dactive contour segmentation of anatomical structures:significantly improved efficiency and reliability. Neuroimage2006;31:1116-28.
16. Antiga L, Steinman DA. Robust and objective decompositionand mapping of bifurcating vessels. IEEE Trans Med Imaging2004;23:704-13.
17. Kontopodis N, Metaxa E, Gionis M, et al. Discrepancies indetermination of abdominal aortic aneurysms maximumdiameter and growth rate, using axial and orthogonalcomputed tomography measurements. Eur J Radio!2013;82:1398-403.
18. Valant AZ, Ziberna L, Papaharilaou Y, et al. The influence oftemperature on rheological properties of blood mixtureswith different volume expanders-implications in numericalarterial hemodynamics simulations. Rheologica Acta2011;50:389-402.
19. Fraser KH, Meagher S, Blake JR, et al. Characterization of anabdominal aortic velocity waveform in patients with abdominal aortic aneurysm. Ultrasound Med Biol 2008;34:73-80.
20. Himburg HA, Grzybowski DM, Hazel AL, et al. Spatial comparison between wall shear stress measures and porcinearterial endothelial permeability. Am J Physiol Heart CircPhysiol 2004;286:Hl 916-1922.
21. Ayachit U. The ParaView Guide: a Parallel VisualizationApplication. Clifton Park, NY: Kitware, Inc, 2015.
22. Truijers M, Pol JA, Schultzekool LJ, et al. Wall stress analysisin small asymptomatic, symptomatic and ruptured abdominal aortic aneurysms. Eur J Vase Endovasc Surg 2007;33:401-7.
23. Garcfa-Garcia F, Metaxa E, Christodoulidis S, et al. Prognosisof Abdominal Aortic Aneurysms Merging Clinical, Morphometric, Biomechanical and Texture Information: a First Ma -chine Learning-enabled Approach. Thessaloniki, Greece:IEEE CBMS, 2017.
24. Shang EK, Nathan DP, Woo EY, et al. Local wall thickness in finite element models improves prediction of abdominalaortic aneurysm growth. J Vase Surg 2015;61:217-23.
25. Raut SS, Jana A, De Oliveira V, et al. The importance ofpatient-specific regionally varying wall thickness in abdom -inal aortic aneurysm biomechanics. J Biomech Eng2013;135:81010.
26. Shum J, DiMartino ES, Goldhamme A, et al. Semiautomaticvessel wall detection and quantification of wall thickness in computed tomography images of human abdominal aorticaneurysms. Med Phys 2010;37:638-48.