Author + information
- Received January 20, 2017
- Revision received April 10, 2017
- Accepted April 20, 2017
- Published online July 3, 2017.
- Zhen Qian, PhDa,b,∗ (, )
- Kan Wang, PhDc,d,
- Shizhen Liu, MD, PhDa,b,
- Xiao Zhou, MD, PhDe,
- Vivek Rajagopal, MDb,
- Christopher Meduri, MDb,
- James R. Kauten, MDb,
- Yung-Hang Chang, MSc,d,
- Changsheng Wu, BSd,f,
- Chuck Zhang, PhDc,d,
- Ben Wang, PhDc,d,f and
- Mani A. Vannan, MBBSa
- aDepartment of Cardiovascular Imaging, Piedmont Heart Institute, Atlanta, Georgia
- bMarcus Heart Valve Center, Piedmont Heart Institute, Atlanta, Georgia
- cH. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia
- dGeorgia Tech Manufacturing Institute, Georgia Institute of Technology, Atlanta, Georgia
- eDepartment of Cardiology, Chinese PLA General Hospital, Beijing, China
- fSchool of Materials Science and Engineering, Georgia Institute of Technology, Atlanta, Georgia
- ↵∗Address for correspondence:
Dr. Zhen Qian, Department of Cardiovascular Imaging, Piedmont Heart Institute, 95 Collier Road, STE 2075, Atlanta, Georgia 30309.
Objectives This study aimed to develop a procedure simulation platform for in vitro transcatheter aortic valve replacement (TAVR) using patient-specific 3-dimensional (3D) printed tissue-mimicking phantoms. We investigated the feasibility of using these 3D printed phantoms to quantitatively predict the occurrence, severity, and location of any degree of post-TAVR paravalvular leaks (PVL).
Background We have previously shown that metamaterial 3D printing technique can be used to create patient-specific phantoms that mimic the mechanical properties of biological tissue. This may have applications in procedural planning for cardiovascular interventions.
Methods This retrospective study looked at 18 patients who underwent TAVR. Patient-specific aortic root phantoms were created using the tissue-mimicking 3D printing technique using pre-TAVR computed tomography. The CoreValve (self-expanding valve) prostheses were deployed in the phantoms to simulate the TAVR procedure, from which post-TAVR aortic root strain was quantified in vitro. A novel index, the annular bulge index, was measured to assess the post-TAVR annular strain unevenness in the phantoms. We tested the comparative predictive value of the bulge index and other known predictors of post-TAVR PVL.
Results The maximum annular bulge index was significantly different among patient subgroups that had no PVL, trace-to-mild PVL, and moderate-to-severe PVL (p = 0.001). Compared with other known PVL predictors, bulge index was the only significant predictor of moderate-severe PVL (area under the curve = 95%; p < 0.0001). Also, in 12 patients with post-TAVR PVL, the annular bulge index predicted the major PVL location in 9 patients (accuracy = 75%).
Conclusions In this proof-of-concept study, we have demonstrated the feasibility of using 3D printed tissue-mimicking phantoms to quantitatively assess the post-TAVR aortic root strain in vitro. A novel indicator of the post-TAVR annular strain unevenness, the annular bulge index, outperformed the other established variables and achieved a high level of accuracy in predicting post-TAVR PVL, in terms of its occurrence, severity, and location.
- annular bulge index
- aortic valve calcification
- balloon post-dilation
- computed tomography
Transcatheter aortic valve replacement (TAVR) has emerged as a treatment option for patients with severe symptomatic aortic stenosis at high risk for surgical aortic valve replacement (1). Computed tomography (CT) and 3-dimensional (3D) echocardiography are routinely used to characterize the aortic root anatomy and size the annulus (2). However, due to the sutureless nature of the TAVR procedure, the occurrence of mild or more paravalvular leak (PVL) after TAVR is higher than after surgical aortic valve replacement, affecting 26% to 67% of patients (3,4). Multicenter studies have shown that moderate-to-severe PVL is an independent risk factor for increased short- and long-term mortality (5,6). Although a number of PVL predictors have been proposed (7), there is no broad consensus on an optimal strategy in patient or prosthesis selection to reduce PVL.
The 3D printing process has been proposed as a fast and cost-effective way to accurately reproduce patient-specific anatomies for education, training (8), and pre-procedural planning (9). Multimaterial printing using colorful soft and rigid materials creates realistic visual and tactile experiences (10), and some day they might be used for 3D-printing-based functional evaluations (11,12). Recently, our group developed a novel metamaterial printing technique using a conventional PolyJet 3D printer (Stratasys, Eden Prairie, Minnesota) and commercial printing materials to produce models with nonlinear and anisotropic mechanical properties, which were comparable to biological tissues, specifically the aortic root (13,14).
We hypothesized that the uneven distribution of strain in the annulus during TAVR contributed to the occurrence of PVL. Thus, in this study, we aimed to develop a procedure simulation platform for in vitro TAVR implantation using 3D printed tissue-mimicking phantoms and CT-based strain quantification techniques and to investigate the feasibility of applying this framework to predict the occurrence, severity, and location of post-TAVR PVL.
This was a retrospective, single-center, observational study approved by the Institutional Review Board of Piedmont Healthcare. We included 22 patients who underwent clinically indicated TAVR with a CoreValve system (self-expanding valve) (Medtronic, Minneapolis, Minnesota) between April 2014 and September 2015. The patients were selected using stratified random sampling, in which 7 to 8 patients were randomly selected in the none, trace-to-mild, and moderate-to-severe groups that constituted a representative spectrum of different degrees of post-TAVR PVL. To reduce operator-induced variability in the TAVR simulation, we did not include patients who received balloon-expandable valves in this study. Before the TAVR procedure, all patients received a contrast-enhanced cardiac CT scan. Prosthesis size was determined by the CT-derived annular diameter as per standard recommendations (15). During the TAVR procedure, valve implantation was performed under the guidance of fluoroscopy and transesophageal echocardiography (TEE). Four of 22 patients were excluded from the study. Two of the excluded patients had predominant subannular deployment of the first valve, requiring a second prosthesis (valve-in-valve); and 2 others had damage to the 3D-printed phantoms due to technical issues (the explanation is in the discussion section). The initial positioning and anchoring of the self-expanding valve system was optimal and successful in the remaining 18 patients. However, TEE revealed that 7 patients had moderate-to-severe PVL after the initial valve deployment, which required post-deployment balloon dilation in an attempt to reduce PVL. TEE post–balloon dilation showed the PVL in 3 of these patients were reduced to trace or mild, and in the other 4, the PVL degrees remained unchanged.
Pre-TAVR CT imaging and image analysis
The pre-procedural contrast-enhanced CT scan was performed with a 320-detector row CT scanner (Aquilion ONE, Toshiba Medical Systems, Otawara, Japan), using an institutional TAVR CT protocol (Online Appendix). Annular diameter was measured at the phase of peak aortic opening. Annular ellipticity was calculated by dividing the maximum diameter by the minimum diameter. Total and regional calcium volumes were quantified in the aortic root, the left ventricular outflow tract, and the annular region (Online Appendix).
TEE evaluation of PVL
PVL was evaluated by TEE at least 10 min later after valve deployment as recommended by the valve manufacturer. PVL was graded as none, trace, mild, moderate, or severe, based on the VARC (Valve Academic Research Consortium)-2 recommendations (16).
3D modeling and 3D printing
The pre-TAVR CT images taken at peak aortic valve opening were identified and used to create the 3D model of the aortic root. Research software (CT Auto Valve, Siemens Corporate Technology, Princeton, New Jersey) was used to semi-automatically segment the images and produce a 3D model of the aortic root. Proprietary software was used to refine the 3D model and empirically add a 2.0-mm wall thickness to the aortic root and a 0.5-mm thickness to the leaflets. Calcified lesions were extracted and converted into a 3D mesh model. Based on the metamaterial configurations previously described (13), sinusoidal fibers were embedded in the 3D model of the aortic wall to achieve strain-stiffening properties comparable to human aortic tissues (Online Appendix). Finally, the 3D models were converted to the Stereolithography format and exported to a multimaterial 3D printer (Stratasys) for printing (Figure 1).
TAVR implantation in vitro
For each patient, according to the size and model used in the clinical procedure, the same self-expanding valve prosthesis was selected and manually implanted in the 3D printed phantom in vitro. The prosthetic valve was carefully deployed to the same depth as in the clinical procedure and manually adjusted to ensure optimal orientation and apposition in the phantom (Figure 2). Because the self-expanding valve is a shape memory device made of nitinol, the phantom and the implanted prosthesis were submerged in 37°C water to ensure the full expansion of the valve as in the in vivo environment.
CT-based strain quantification and bulge detection
To quantify the post-TAVR aortic root strain distribution in the 3D-printed phantom, small radiopaque beads were attached to the phantom to serve as landmarks. The positions of the landmarks before and after the in vitro TAVR implantation were obtained by performing 2 CT scans, based on which the circumferential strain distribution was quantified (Online Appendix). At the annular level, maximum, mean, and minimum strain values were calculated in each phantom.
A bulge detector was designed to detect the low-high-low strain pattern along the phantom’s annulus after the in vitro implantation (Online Appendix). The width of the detector’s positive peak was set to equal the circular angle between 2 adjacent struts at the ventricular end of the self-expanding valve (Figure 3). The bulge index was calculated by convolving the annular strain with the bulge detector, and in each phantom, the maximum bulge index was reported. Additionally, to assess the effect of varying the detector scale, different detector widths were also tested.
Continuous variables were reported as mean ± SD. Statistical significance was defined as p < 0.05. Patients were divided into 3 subgroups based on the TEE-indicated PVL degrees after the initial valve deployment and after the balloon post-dilation. The subgroups were defined as follows: 1) no PVL; 2) trace-to-mild PVL; and 3) moderate-to-severe PVL. One-way analysis of variance and Kruskal-Wallis test were used to test the difference between the subgroups, when Levene test was negative and positive, respectively. Mann-Whitney-Wilcoxon test and receiver-operating characteristic (ROC) analysis were performed using greater-than-or-equal-to moderate PVL as the classification variable. The area under the curve (AUC) was calculated, and the cutoff value was determined by searching for the highest combination of sensitivity and specificity. Net reclassification improvement was also calculated. Statistical analysis was performed using Medcalc version 16.4.3 (Medcalc Software, Ostend, Belgium).
This study included 8 men and 10 women with a mean age of 79.6 ± 8.9 years. TAVR was performed through the transfemoral approach in 12 patients (66.7%) and through the subclavian approach in 6 patients (33.3%). The following self-expanding valve sizes were implanted: 23 mm in 1 patient; 26 mm in 6 patients; 29 mm in 4 patients; 31 mm in 4 patients; CoreValve Evolut R (second generation self-expanding transcatheter aortic valve) 26 mm in 1 patient; and second generation self-expanding transcatheter aortic valve 29 mm in 2 patients. The average implantation depth, which was defined as the distance from the lower ventricular end of the prosthesis to the aortic annulus, was 4.8 ± 0.6 mm. After valve deployment, TEE revealed that 6 patients (33.3%) had no PVL, 5 patients (27.8%) had trace-to-mild PVL, and 7 patients (38.9%) had moderate-to-severe PVL. In the 7 patients with significant PVL, post-deployment balloon dilation was attempted to reduce PVL. Three patients’ (16.7%) PVL was reduced to trace or mild, whereas the others’ (22.2%) PVL did not improve. Table 1 shows the detailed patient characteristics.
3D modeling and 3D printing
The semi-automated 3D modeling of the aortic root took approximately 5 to 10 min. The next step to refine the 3D model and generate the Stereolithography file was automated and took <5 min. The 3D printer was able to read in Stereolithography files data from up to 10 patients in 1 batch, and it took a total of 9 to 10 h to print these 10 3D phantoms simultaneously. Post-print processing, such as removing the support materials and attaching the radiopaque beads, took about 45 min for each phantom. The cost of the printing materials in each phantom was approximately $150 to $200.
Annular strain and bulge quantification
Each phantom’s strain distribution showed a distinctive pattern (Figure 4). At the annular level, the maximum, mean, and minimum circumferential strain was 15.7 ± 4.3%, 8.1 ± 3.6%, 1.3 ± 4.1%, respectively. The strain was not statistically different in the 3 PVL subgroups (Table 2).
As shown in Table 3, the maximum annular bulge index was significantly different among the 3 PVL subgroups (p = 0.047) after valve deployment, with higher bulge index being associated with the higher degree of PVL. Pairwise comparison showed that the bulge index in the moderate-to-severe subgroup was significantly higher than that in the non-PVL subgroup only (Figure 5). Balloon dilation was done after initial deployment for greater than mild PVL in 7 patients. Three of these patients were reassigned to the trace-to-mild subgroup. Similar to pre-balloon dilation, the maximum annular bulge index was significantly different among the 3 reclassified subgroups (p = 0.001). But, pairwise comparison showed that the bulge index in the moderate-to-severe PVL group was now, significantly higher than in the other 2 subgroups.
Among other quantitative parameters, only the annular calcium volume was significantly different among the subgroups (p = 0.048) (Table 3) before balloon dilation. There was a trend for higher annular calcium volume to be associated with the higher degree of PVL, but without significant pairwise difference. However, post–balloon dilation none of these other parameters were significantly different among the subgroups (Table 3).
Prediction of PVL
As shown in Table 4, annular calcification was the best predictor of moderate-to-severe PVL (ROC AUC = 83%; 95% confidence interval [CI]: 58% to 96%; p < 0.001) after initial deployment. Bulge index was a significant but less accurate predictor (AUC = 77%; 95% CI: 51% to 93%; p = 0.04). However, no other variables were predictive of PVL. Post-balloon expansion annular calcium lost its predictive power. Instead, bulge index became the only significant predictor of moderate-to-severe PVL (AUC = 95%; 95% CI: 73% to 99.9%; p < 0.0001), and it achieved a net reclassification improvement of 25% over annular calcium. In Figure 6, the ROC comparison shows that post-balloon dilation, the bulge index has a higher AUC than the annular calcium volume, with a nonsignificant trend toward more accurate prediction of PVL (p = 0.20).
In the 3 patients whose PVL decreased from moderate to severe to trace to mild after balloon dilation, the mean bulge index was 4.8 ± 2.2% (individually 3.0%, 4.2%, 7.3%) and the annular calcium volume was 165 ± 71 ml (individually 103 ml, 151 ml, 242 ml). In the other 4 patients in whom the PVL did not improve after balloon dilation, the mean bulge index of 9.1 ± 2.5% (individually 5.7%, 9.9%, 11.8%, 9.0%) was higher. But, the annular calcium volume of 149 ± 78 ml (individually 156 ml, 42 ml, 228 ml, 169 ml) was lower than in the 3 who showed improvement.
Figure 7 shows the locations of PVL in the 12 patients who had any degree of PVL post-TAVR and the sites of the maximum annular bulge index. The latter predicted the location of the dominant PVL in 9 patients (accuracy = 75%). In patients #1 and #5, who had multiple PVL sites, the second largest bulge index position predicted the dominant PVL location, whereas the maximum bulge index predicted a minor PVL site. In patient #6, the annular strain distribution showed multiple high bulges (indicated by the warm color) besides the maximum bulge site, and only 1 of them predicted PVL.
In this proof-of-concept study, we have demonstrated the feasibility of using 3D-printed tissue-mimicking phantoms to quantitatively assess the post-TAVR aortic root strain in vitro. Annular strain unevenness, which we formulated as the annular bulge index, outperformed other established variables and was accurate in predicting post-TAVR PVL, in terms of the occurrence, severity, and location.
To our knowledge, this study is the first attempt to use tissue-mimicking 3D printing as a quantitative tool to predict TAVR outcomes. Compared with the traditional uses of 3D printing in medicine, such as reproducing patient-specific anatomies (8,9), the use of 3D printing as a quantitative tool to study pathophysiology is relatively new. Moreover, we used a unique approach combining commercially available photopolymers to design a metamaterial that would mimic the mechanical properties of the human aortic tissue when subjected to deforming pressure. Computer-based numerical analysis, such as finite element analysis, has also been used to quantitatively simulate TAVR implantation (17). However, these numerical approaches are based on some assumptions about the material properties and the contact/friction constraints of the aortic tissue and the prosthesis. In our approach, the actual prosthetic valves were used for in vitro deployment, which greatly improved the fidelity of the simulation. Additionally, the 3D-printed phantom provided a unique interactive platform to the users with both visual and tactile experiences, which are critical for simulation of the procedure, but are not available in numerical simulations. Lastly, the relatively fast and inexpensive 3D printing method proposed in this study may provide a better and more practical alternative to computer-based simulations. This approach of 3D printing of patient-specific phantoms using tissue-mimicking materials may potentially help with selecting the optimal valve type/size, and optimizing the deployment techniques including post-dilation of the valve particularly in patients where conventional CT predicts high risk of PVL.
Balloon post-dilation of the prosthesis is often performed post-deployment in patients who had significant PVL after the initial valve deployment. Whether this fixes the PVL is hard to predict, and the mechanism of such a fix was not fully understood (18). Predictors of PVL include landing zone calcification, elliptical shape of the annulus, implantation depth, and size of the prosthesis (6,19). In this study, annular strain was not associated with the presence or the degree of PVL. This may be because we used standard pre-procedural imaging methods to size the valves, which resulted in similar annular strain in all 3 groups. Instead, the bulge index, which is a measure of annular unevenness (areas of low-high-low strain pattern), was predictive of PVL both after initial deployment (pre-balloon) and post-balloon. Additionally, the annular calcium volume was predictive of PVL only after the initial valve deployment, but its predictive value was lost if post-dilation was performed. A possible explanation to this is that the bulge index is a mechanical descriptor of the strain-stress mismatch in the post-TAVR annulus, and therefore it is more directly associated with PVL than morphological descriptors, such as the annular calcium volume. In addition, balloon dilation likely worked as a fine-tuning tool that smoothed the strain mismatch caused by calcium and adjusted the contact between the prosthesis and the aortic root, and thus improved the sealing of the annulus.
Furthermore, we found that a bulge detector scale that equaled the interstrut angle of the self-expanding valve at the ventricular end (24° for ≥26-mm and 30° for 23-mm valve sizes) performed the best with regard to the prediction of greater-than-or-equal-to moderate PVL before and after balloon dilation (Figure 8). On the contrary, smaller and larger scale bulge detectors were not as good for predicting significant PVL. Perhaps the random distribution of the severity and location of annular calcium, which creates various scales of strain unevenness, results in imperfect annular sealing and significant PVL. Balloon dilation likely smoothed out these areas of large annular unevenness in a global fashion. However, it may not as effectively fix focal PVL that is caused by small scale of strain unevenness that extends between 2 adjacent struts. Thus, annular calcium volume was superior to the bulge detector pre-balloon to predict significant PVL. However, if the bulge persists between 2 consecutive struts even after balloon dilation, then there is likely to be significant residual PVL. This hypothesis is exemplified by Patient #5 in Figure 7 where the pre-balloon PVL was located at the 12 o’clock position (Online Figure 1) within an area of large annular strain but with an angular extent that was larger than 24° and, hence, was not seen as a hot area in the bulge index map. This PVL disappeared post-balloon as we would expect from the hypothesis that the balloon dilation evens out large bulge areas. Additionally, the final residual mild PVL was now located at the 5 and 7 o’clock positions at the sites of the bulges with angular extents close to 24°.
In this study, we found that the annular calcification in TAVR was an important risk factor of PVL, but not all incidents of them actually lead to PVL. As shown in Table 4, for the prediction of greater-than-or-equal-to moderate PVL after balloon dilation, the specificity of annular calcification was only 50%, whereas the specificity of the bulge index was 79%. We propose that if valve sizing and deployment is optimal, then the amount of annular calcification is a necessary but not sufficient condition for the occurrence of PVL. Instead, the shape, size, and location of the annular calcification (20) may be instrumental in causing focal uneven strain distribution (high bulge index) resulting in significant PVL. The 3D-printed phantom provides a practical in vitro way to quantitatively assess the distribution of post-TAVR annular strain. This may improve our understanding of the role of the annular calcification in the genesis of PVL, and it may be extendable to other transcatheter valve therapies.
Although this feasibility study shows promise, further research is warranted to optimize and standardize the printing and simulation protocols. For example, 2 patients were excluded from the study because the 3D-printed phantoms were damaged during the CT strain imaging. This might have occurred because the phantoms with the valve implant were submerged in the 37°C water for too long (approximately 15 min longer than the undamaged ones). This may have led to material fatigue and creep. A better understanding of the printing material’s behavior in a warm-water environment is desirable, and the length of the warm bath has to be optimized and standardized to ensure optimal and reproducible results.
There is room for improvement in the performance of the bulge detector. For example, as shown in Figure 7, most patients had multiple sites of high bulge index. Although the maximum bulge index predicted the major leak site, it was not clear whether the other regions of high bulges would lead to multiple sites of PVL. Further research is needed to achieve a more comprehensive assessment. More importantly, the formulation of the bulge index in this study was specifically designed for the self-expanding valve prosthesis. To extend this tool to other prosthetic valves, such as the balloon-expandable valves and the emerging valves with sealing cuffs (5), we expect to design new bulge detectors.
Finally, this was a retrospective study with a relatively small patient cohort that consisted of subgroups with different degrees of PVL outcomes, rather than a prospective study of consecutive patients, and, therefore, was subject to sampling biases. Additionally, this study included both the first- and second-generation self-expandable valves, which are found to have different rates of PVL in practice. Thus, caution must be exercised in interpreting the results, and further research with a larger prospective cohort is needed.
The annular bulge index is a novel indicator of the post-TAVR annular strain unevenness. It can be quantified by in vitro TAVR simulation on a patient-specific 3D-printed phantom, using unique tissue-mimicking metamaterials. This bulge index outperformed established variables and achieved a high degree of accuracy in predicting the occurrence, severity, and location of post-TAVR PVL. Thus, it may be feasible to perform procedural simulations on a 3D-printed phantom for pre-TAVR planning, especially in those who are at high-risk for post-TAVR PVL. This may refine the current approach for the selection of valve type/size and potentially reduce the rate of post-TAVR PVL.
COMPETENCY IN MEDICAL KNOWLEDGE: Moderate-to-severe post-TAVR PVL is an independent risk factor of increased short- and long-term post-TAVR mortality. It is feasible to quantitatively assess the post-TAVR aortic root strain in vitro using tissue-mimicking 3D-printed phantoms. A novel indicator of the post-TAVR annular strain unevenness, the annular bulge index, has been developed to provide new pathophysiological insight into PVL formation, and it may have an advantage over the other established variables in predicting post-TAVR PVL.
TRANSLATIONAL OUTLOOK 1: It may be appropriate to perform procedural simulations on a 3D-printed tissue-mimicking phantom and gather relevant pathophysiological information for pre-TAVR planning among certain high-risk patient populations.
TRANSLATIONAL OUTLOOK 2: Although this feasibility study shows promise, further research with a larger prospective cohort is needed to investigate whether the information gathered from the 3D-printed patient-specific phantom would refine the selection of current TAVR prosthesis for a given patient and improve the design of future valves to reduce the rate of PVL.
The authors are grateful to Medtronic for providing the CoreValve samples and Siemens Corporate Technology for providing the CT Auto Valve research software.
For a supplemental figure and material, please see the online version of this article.
Dr. Rajagopal has received speaking honoraria from Edwards Lifesciences and Medtronic; and has served as a proctor for Edwards Lifesciences and Medtronic. Dr. Meduri has received consulting honoraria from Boston Scientific and Medtronic; has received an educational grant from Medtronic; has served on the Advisory Board of Boston Scientific; and has served as a proctor for Boston Scientific and Medtronic. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- area under the curve
- confidence interval
- computed tomography
- paravalvular leak
- receiver-operating characteristic
- transcatheter aortic valve replacement
- transesophageal echocardiography
- Received January 20, 2017.
- Revision received April 10, 2017.
- Accepted April 20, 2017.
- 2017 American College of Cardiology Foundation
- Webb J.G.,
- Wood D.A.
- Manoharan G.,
- Walton A.S.,
- Brecker S.J.,
- et al.
- Généreux P.,
- Head S.J.,
- Hahn R.,
- et al.
- Unbehaun A.,
- Pasic M.,
- Dreysse S.,
- et al.
- Valverde I.,
- Gomez G.,
- Suarez-Mejias C.,
- et al.
- Maragiannis D.,
- Jackson M.S.,
- Igo S.R.,
- et al.
- Wang K.,
- Zhao Y.,
- Chang Y.,
- et al.
- Wang K.,
- Wu C.,
- Qian Z.,
- et al.
- Kasel A.M.,
- Cassese S.,
- Bleiziffer S.,
- et al.
- Kappetein A.P.,
- Head S.J.,
- Généreux P.,
- et al.
- Wang Q.,
- Primiano C.,
- McKay R.,
- McKay R.,
- Kodali S.,
- Sun W.
- Nombela-Franco L.,
- Rodés-Cabau J.,
- DeLarochellière R.,
- et al.
- Detaint D.,
- Lepage L.,
- Himbert D.,
- et al.
- Azzalini L.,
- Ghoshhajra B.B.,
- Elmariah S.,
- et al.