Article Text

Download PDFPDF

Original research
Increased focal internal carotid artery angulation in patients with posterior communicating artery aneurysms
  1. Richard Rosato,
  2. Gabriela Comptdaer,
  3. Ryan Mulligan,
  4. Jeffrey M Breton,
  5. Emal Lesha,
  6. Alexandra Lauric,
  7. Adel M Malek
  1. Cerebrovascular and Endovascular Division, Department of Neurosurgery, Tufts Medical Center and Tufts University School of Medicine, Boston, Massachusetts, USA
  1. Correspondence to Dr Adel M Malek, Cerebrovascular and Endovascular Division, Department of Neurosurgery, Tufts Medical Center, Boston, MA 02111, USA; amalek{at}tuftsmedicalcenter.org

Abstract

Background Aneurysms at the posterior communicating artery (PCOM) origin represent the most common location on the internal carotid artery (ICA), and are associated with greater recurrence following endovascular treatment. We evaluate the association between ICA angulation in three-dimensional (3D) space and PCOM aneurysmal development, using high-resolution 3D rotational angiography (3DRA) studies.

Methods 3DRA datasets were evaluated in 70 patients with PCOM aneurysms, 31 non-aneurysmal contralateral, and 86 healthy controls (187 total). The local angle formed by upstream and downstream ICA segments at the PCOM origin, αICA@PCOM, was measured using 3DRA multiplanar reconstruction. Computational fluid dynamics (CFD) analysis was performed on parametric and patient-based models.

Results αICA@PCOM was significantly larger in aneurysm-bearing ICA segments (68.14±11.91°) compared with non-aneurysmal contralateral (57.17±10.76°, p<0.001) and healthy controls (48.13±13.68°, p<0.001). A discriminant threshold αICA@PCOM value of 61° (87% specificity, 80% sensitivity) was established (area under the curve (AUC)=0.88). Ruptured PCOM aneurysms had a significantly larger αICA@PCOM compared to unruptured (72.65±15.16° vs 66.35±9.94°, p=0.04). In parametric and patient-based CFD analysis, a large αICA@PCOM induces high focal pressure at the PCOM origin, relatively low wall shear stress (WSS), and high proximal WSS spatial gradients (WSSG).

Conclusion ICA angulation at PCOM origin is significantly higher in vessels harboring PCOM aneurysms compared with contralateral and healthy ICAs. This sharper bend in the ICA leads to high focal pressure at the aneurysm neck, low focal WSS and high proximal WSSG. These findings underline the importance of morphological ICA variations and the likelihood of PCOM aneurysm, an association which can inform clinical decisions and may serve in predictive analytics.

  • aneurysm
  • angiography

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Cerebral bifurcations and high curvature arterial segments are preferred sites for intracranial aneurysm formation.1 Intracranial aneurysms—pathologic focal dilatations of the cerebrovasculature—occur in the general population with a frequency of 3.6–6%. The origin of the posterior communicating artery (PCOM) is among the most common locations for aneurysm formation, accounting for roughly 30–35% of all intracranial aneurysms, and 50% of all internal carotid artery (ICA) aneurysms.2 These aneurysms occur on the communication segment of the ICA, and can be classified as either sidewall or bifurcation aneurysms,3 depending on the size of the PCOM, which can vary from barely detectable to very prominent. Considerable variations in the anatomy of the PCOM complex has surgical and endovascular implications, especially as PCOM aneurysms have a high recurrence rate (37.2%) post-endovascular treatment.4 Consequently, understanding the morphology of the communicating segment of the ICA associated with the presence of PCOM aneurysms is clinically meaningful.

The mechanisms for pathogenesis and the development of intracranial aneurysms remain unclear, but vessel morphology and hemodynamic patterns play a role and have been shown to be interrelated. Bifurcation aneurysms were associated with wide angular morphology at the aneurysm site,5 6 whereas sidewall aneurysms were associated with high vessel curvature,7 8

In contrast to previous work, this current study has a larger sample size of high-resolution three-dimensional (3D) volumetric data. The focal ICA angle at the PCOM origin (αICA@PCOM) is evaluated in aneurysmal, contralateral, and healthy controls. The large sample data allow for subgroup analysis such as gender, ruptured/unruptured status, and sidewall/bifurcation status. Computational fluid dynamic (CFD) analysis is performed on parametric and patient-derived data in order to understand the hemodynamic implications of the local ICA angulation at the PCOM location.

Methods

Patient selection

Consecutive high-resolution 3D angiographic data of patients undergoing cerebral angiography between 2009 and 2016 were reviewed for this study. All angiographic studies satisfying the following criteria were included: patients with PCOM aneurysms, and imaging of sufficient quality to permit accurate image processing and morphological analysis. When available, bilateral data of the non-aneurysmal contralateral were included in the study. Consequently, volumetric data from 70 PCOM aneurysms were available. The non-aneurysmal contralateral was available in 31 patients. In addition, 86 datasets from healthy patients with no aneurysms were included. This resulted in a total of 187 angiographic samples available for analysis.

This research was approved by Tufts Health Sciences Campus Institutional Review Board (IRB# 9035). Patient consent was waived because of the retrospective nature of the study not impacting patient clinical care.

Angiography and data processing

Three-dimensional catheter-based cerebral angiograms (3DRA) were obtained using a Siemens Artis (Malvern, PA) calibrated biplane system, with volume reconstruction using the available Siemens Leonardo clinical software package, to yield a 3D volumetric dataset. Volumetric datasets, including aneurysm and parent vessel, were analyzed using Amira 5.3 (FEI, Hillsboro, OR).

Posterior communicating artery angle

The ICA surface and the ICA skeleton (centerline) were generated in Amira from the catheter 3D rotational angiography volumetric datasets. The 3D vessel skeleton was created for the entire ICA using the Amira auto-skeleton tool. Using the local vessel centerline as a guideline, multiplanar reconstruction through the volume was performed at the site of the PCOM in order to ensure the plane was close to the center of the vessels, and to capture the local bend and measure the αICA@PCOM angle in two-dimension (2D) following the ICA bend at the PCOM location (figure 1). αICA@PCOM angles were measured on the 2D cuts as the angle between the ophthalmic ICA (ICA proximal to the PCOM) and the communicating ICA (ICA distal to the PCOM). As shown in figure 1E,F, angles were measured in the direction of the flow, using the convention set by Rossitti and Lofgren.9 Specifically, angles represent the deviation from the direction of the flow in the proximal segment.5 9 10 This measuring convention is in contrast to previous research reporting on the PCOM angular morphology,11 12 but it is consistent with previous studies evaluating angular deviations.5 6 9 10

Figure 1

Angle measurement in a patient with unilateral PCOM aneurysm. (A) Lateral DSA of the non-aneurysmal contralateral side. (B) Lateral DSA of the aneurysmal side. (C–D) Volume rendering projections of bilateral 3DRA data in a patient with unilateral PCOM aneurysm. (E) Measurement detail for the non-aneurysmal contralateral. αICA@PCOM angles were measured on two-dimensional longitudinal cuts through the three-dimensional volume. Vessel centerline is shown in red. (F) Measurement detail for the aneurysmal side. PCOM, posterior communicating artery; AChA, anterior choroidal artery; αICA@PCOM, focal internal carotid artery angle at the posterior communicating artery origin; DSA, digital subtraction angiography; 3DRA, three-dimensional rotational angiography

All measurements were performed independently by two trained operators. The inter-operator correlation for the set of 187 measurements was 0.85, characterized as good data reliability.13

Statistical analysis

JMP statistical software (Version Pro 14.0, SAS Institute, Cary, NC) was used to evaluate the performance of all parameters in discriminating between aneurysmal and control PCOM datasets. Statistical significance was assumed for p<0.05. All variables were tested independently using the Student t-test analysis. Contralateral analysis was performed using matched-pair analysis. Receiver operating characteristics curve (ROC) was computed, and prediction accuracy was evaluated using the area under the curve (AUC) index. Optimal threshold values were determined using ROC statistics.

Patient-derived models

Ten aneurysmal and 10 healthy volumetric data were randomly selected for CFD analysis. Following vessel segmentation in Amira, the aneurysms were digitally removed by local Laplacian smoothing,14 and the parent vessels were reconstructed in MeshLab version 1.3.1 (ISTI-CNR, Pisa, Italy). Laplacian filtering tends to shrink the surface being smoothed. A high number of smoothing iterations was applied on an area constrained to the aneurysm dome, which resulted in the removal of the aneurysm while preserving the surrounding vasculature. The segment proximal to the site of the aneurysm was kept sufficiently long (at least five times the diameter of the parent vessel) in order to enable the laminar flow to become fully developed and avoid entrance effects.

Parametric models

Parametric models of the communicating segment of the ICA were constructed in order to account for the morphological configurations observed in patient datasets with respect to the PCOM angle (figure 2). All models were created using Space Claim (Ansys 19.2, Ansys Inc, Canonsburg, PA).

Figure 2

CFD simulations on parametric models. Because of the straight distal segment, the initial model was labeled as 0°. Three subsequent models were created adding a second bend at inclinations of 20°, 40° and 60°, respectively. (A) Pressure, (B) WSS, and (C) WSSG variations with increase in αICA@PCOM angles. (D) Detail into evaluating the pressure, WSS and WSSG ratios: plotting of the three ratio parameters. αICA@PCOM, focal internal carotid artery angle at the posterior communicating artery origin; CFD, computational fluid dynamics; WSS, wall shear stress; WSSG, WSS gradient.

The initial model was created with a diameter of 5 mm and consisted of a tubular structure having a 40 mm straight initial segment followed by a 30 mm segment at 50° vessel deviation in the direction of the flow. Because of the straight distal segment, this initial model was labeled as the 0° model in figure 2. Three subsequent models were created adding a second bend with an αICA@PCOM angle of 20°, 40° and 60°, respectively (figure 2).

Computational fluid dynamics

The segmented surfaces were used to create dynamic tetrahedral/prism meshes with boundary layer enhancement. The final mesh was created with a vessel entrance length several times the diameter of the parent vessel using ICEM CFD meshing (Ansys 19.2, Ansys Inc, Canonsburg, PA) (figure 1C).

CFD simulations were performed using Ansys Fluent 19.2 (Ansys Inc, Canonsburg, PA), by modeling steady state laminar flow with a Carreau non-Newtonian profile to better approximate the viscosity and flow of blood.15 In order to account for variability in the size, shape and length of the parent vessel, flow scaling factors were determined for each model to ensure ideal parent vessel wall shear stress (WSS) conditions of approximately 1.5 Pa at proximal ICA segments. Subsequently, steady flow was applied at the inlet according to the resulting scaled velocity, with zero-gauge pressure at the outlets.

Post-processing analysis was performed using EnSight 10.2 (Ansys Inc, Canonsburg, PA). The WSS was evaluated at the site of the aneurysm in aneurysmal models and at the site of the PCOM in healthy models. Normalized CFD parameters (WSS ratio, WSS gradient (WSSG) ratio, and pressure ratio) were defined as the ratio between the focal parameter at the aneurysmal/PCOM site, over the proximal value (figure 2). The proximal ICA radius was used as a relative measure in deciding the position for parameter normalization. In the case of patient models, the proximal position was chosen approximately one radius (~2–3 mm) from the position of the PCOM artery. For parametric models the proximal position was chosen within one radius (~2–3 mm) from the PCOM bend. There were small variations (<1 mm) in the positioning of the proximal area in order to capture the maximal contrast in parameters between the proximal vessel and the PCOM bend.

Results

Patient demographics

A total of 187 3D rotational angiographic volumes were analyzed for this study. There were 70 aneurysmal samples (62 women; 58.40±10.38 years), 31 non-aneurysmal contralateral (29 women; 58.12±10.76 years), and 86 healthy samples (47 women; 55.15±13.09 years). There was no statistically significant difference in age between the three subgroups (table 1). There were 50 unruptured aneurysms (71.4%, 44 women) and 20 ruptured aneurysms (28.6%, 18 women) in the dataset. According to the size of the PCOM relative to the ICA size,3 41 aneurysms (58.5%, 38 women) were labeled as sidewall, and 29 aneurysms (41.5%, 24 women) were labeled as bifurcation. Additional demographic information is available as supplementary data (online supplementary table 1).

Table 1

Univariate and matched pair statistical analysis of the αICA@PCOM angle

Univariate statistical analysis

As shown in table 1, the αICA@PCOM angle was statistically significant in discriminating between aneurysms, contralateral and healthy datasets. PCOM aneurysm presence was associated with a significantly larger αICA@PCOM (68.14±11.91°) compared with the contralateral group (57.17±15.52°, p<0.001) and healthy samples (48.13±13.68°, p<0.001). In addition, the non-aneurysmal contralaterals had significantly larger αICA@PCOM compared with healthy controls (p=0.001). In ROC analysis, αICA@PCOM had an AUC of 0.88 in discriminating between aneurysmal and healthy controls, with an optimal angle threshold of 61° (87% specificity, 80% sensitivity). These results proved to be very robust, and similar findings were reported when the analysis was repeated for female samples, unruptured aneurysms, ruptured aneurysms, sidewall aneurysms, and bifurcation aneurysms (online supplementary table 2).

Ruptured aneurysms had statistically significant larger αICA@PCOM compared with unruptured aneurysms (72.65±15.16° vs 66.35±9.94°, p=0.04). Bifurcation aneurysms were noted to have higher αICA@PCOM compared to sidewall aneurysms, but the difference did not reach statistical significance (70.97±12.90° vs 66.16±10.87°, p=0.10).

Matched pair analysis

Bilateral imaging was available for 31 patients harboring unilateral PCOM aneurysms. Matched pair analysis (table 1), in which each patient acts as its own control, showed that αICA@PCOM was significantly larger in aneurysms compared with the corresponding non-aneurysmal contralateral (69.60±14.62° vs 57.17±15.52°, p<0.001). Analysis was repeated with similar results on female samples (29 pairs), unruptured aneurysms (21 pairs), ruptured aneurysms (10 pairs), sidewall (18 pairs), and bifurcation aneurysms (13 pairs) (online supplementary table 3).

Effect of aneurysm morphology

The average maximal aneurysm size of the dataset was 5.13±3.43 mm. The dataset contained no giant aneurysms, defined as having a size larger than 25 mm. Six of the aneurysms were larger than 10 mm, and 14 aneurysms were larger than 7 mm. The aneurysm size was not statistically different between ruptured and unruptured aneurysms, but ruptured aneurysms tended to be larger compared with unruptured (6.38±4.54 mm vs 4.63±2.75 mm, p=0.06). No correlation was found between aneurysm size and αICA@PCOM. The analysis was repeated for several subgroups (ruptured, unruptured, aneurysms smaller than 7 mm, and aneurysms larger than 7 mm) with similar results. Analysis showed that αICA@PCOM was independent of the aneurysm size, and the ICA at the PCOM angle was significantly larger in aneurysmal samples compared with contralateral and healthy controls, even when larger aneurysms (>7 mm) were excluded from analysis.

The average aneurysm neck size of the dataset was 3.04±1.01 mm. The neck size was not statistically different between ruptured and unruptured aneurysms (3.12±1.24 mm vs 3.01±0.89 mm, p=0.65). No correlation was found between aneurysm neck size and αICA@PCOM The analysis was repeated for the ruptured and unruptured subgroups with similar results.

CFD analysis

CFD analysis on the four parametric models showed that, compared with the initial 0° model, the αICA@PCOM angle induces high focal pressure which increases with the angle of the bend, and is reflected in a high pressure ratio (figure 2D). Increasing the angle of the secondary bend creates a pattern of high WSS proximal to the bend, followed by lower WSS at the elbow location (figure 2). The area covered by low WSS increases with the size of the αICA@PCOM angle, resulting in a decrease in the WSS ratio. These results are reflected in the WSS ratio, which decreases in models with larger αICA@PCOM angles.

Similarly, in all 10 aneurysmal patients evaluated for CFD, the area of the PCOM origin was characterized by relatively high focal local pressure (figure 3A), which was not observed in healthy samples. In these aneurysmal models, the area corresponding to the aneurysm neck is characterized by WSS patterns like those reported in parametrical models (figure 3B). Namely, in contrast to healthy models, the aneurysm site is characterized by high normalized pressure and low normalized WSS with high spatial WSSG variation (figure 3B). Because the healthy models have smaller αICA@PCOM angles, indicative of a milder ICA bend, the same hemodynamic pattern is not visible in these models, which are characterized by small WSS and pressure variations at the bend. Consequently, the WSS ratio and WSSG ratio were significantly lower in aneurysm compared with healthy models (0.63±0.13 vs 0.55±0.09, p=0.02, and 0.75±0.14 vs 1.30±0.59, p=0.01, respectively), indicative of high WSS variation in aneurysmal samples (online supplementary figure 1). In contrast, the pressure ratio was significantly higher in the 10 aneurysmal samples compared with the 10 healthy controls (1.08±0.06 vs 0.95±0.04, p<0.001) (online supplementary figure 1).

Figure 3

(A) Aneurysmal patients evaluated for CFD show relatively high focal pressure at the site of the aneurysm. (B) Compared with healthy patients, higher αICA@PCOM angles in the aneurysmal sample induce (first row) higher focal pressure, (second row) lower focal WSS, and (third row) lower WSSG at the site of the PCOM aneurysm. αICA@PCOM, focal internal carotid artery angle at the posterior communicating artery origin; CFD, computational fluid dynamics; PCOM, posterior communicating artery; WSS, wall shear stress; WSSG, WSS gradient.

Discussion

The PCOM is the most common location for aneurysm occurrence on the ICA. The complexity of the communicating segment morphology poses a challenge in treating these aneurysms due to the high recurrence rate following endovascular treatment. Understanding the morphological patterns associated with the presence of a PCOM aneurysm can inform clinical practice and assist in selecting the most appropriate treatment options.

Vascular angulation was previously evaluated in relation to aneurysm presence,5 6 16 risk of rupture,17 18 and endovascular effects19–22 on vessel morphology. Middle cerebral artery5 and basilar artery bifurcations6 harboring aneurysms had significantly larger bifurcation angles compared with controls. Angulation analysis of aneurysm rupture status showed that the aneurysm inflow angle was significantly more direct into ruptured aneurysms compared with unruptured aneurysms.18 Favorable remodeling and angulation decrease was reported in follow-up studies post-intracranial Y-stenting,19 21 as well as single stent-assisted coiling,20 resulting in a decrease in the area of the impingement zone. Endovascular treatment induces vessel straightening which results in a reduction of pressure, mean velocity and WSS, conditions associated with a reduced risk of aneurysm origination.22

The high curvature of the ICA was previously associated with the presence of downstream ICA aneurysms in the cavernous and supraclinoid regions7 and also with ruptured status.8 Specifically to PCOM, Yu et al 11 analyzed data from nine patients with unilateral PCOM aneurysms, for which bilateral data were available. They reported the angle formed between the ICA and the origin of the PCOM (which can be difficult to visualize depending on technique, caliber and contrast filling) was smaller on the aneurysmal side. In contrast, they found no significant difference in their small sample in the angle formed by the ophthalmic and communicating ICA segments. Analyzing lateral (two-dimensional) views of digital subtraction angiography (DSA), Hu and Wang12 found that the angle formed by the ophthalmic and communicating segments to be smaller in PCOM aneurysms (correspondingly larger projected αICA@PCOM) compared with healthy controls. However, no angular differences were reported between aneurysmal samples and the non-aneurysmal contralaterals. Studies using CT angiography data reported no association between PCOM aneurysm presence and the angular morphology of the ICA.23 24

In contrast to previous studies, our study had a large dataset of 187 samples from high-resolution catheter-based 3D angiography. In contrast to 2D DSA data, 3D models allowed for unrestricted views of the PCOM complex; however, for accurate angle evaluation, measurement of the ICA bend was performed in 3D space guided by the vessel centerline. Bilateral angular differences were confirmed using matched pair statistical analysis. Measurement accuracy was validated by use of multiple operators. CFD analysis was performed on parametric and patient-derived data in order to understand the hemodynamic patterns related to changes in the ICA angle at the PCOM origin.

The αICA@PCOM was significantly larger in aneurysmal samples compared not only with healthy controls (p<0.001, AUC=0.88, optimal threshold 61°), but also with contralateral data (p<0.001). Moreover, non-aneurysmal contralateral samples to PCOM aneurysms had a significantly larger αICA@PCOM angle compared with healthy non-aneurysmal controls (p=0.001). The large data sample allowed for subgroup analysis, and similar results were found when analysis was repeated on females only, unruptured aneurysms, ruptured aneurysms, and sidewall and bifurcation aneurysms. In addition, ruptured aneurysms had a significantly larger αICA@PCOM angle compared with unruptured aneurysms (p=0.04). Baharoglu et al 18 previously proposed the aneurysm inflow angle as a discriminant of rupture in sidewall aneurysms. This parameter was defined as the angle formed between the axis of flow in the proximal parent vessel at the level of the aneurysm neck, and the aneurysm’s main axis from the center of the neck to the tip of the dome.18 Clearly this measure can only be evaluated in aneurysmal samples. In contrast αICA@PCOM captures angular morphological information in the case of both aneurysmal and non-aneurysmal samples focusing on the analysis of parent vessel morphology to elucidate contribution to aneurysmogenesis. In order to determine the hemodynamic conditions that are associated with a high αICA@PCOM angle, the CFD simulations in this study were performed on parametric models without aneurysms and on patient models with digitally removed aneurysms. Consequently, this analysis was not influenced by the aneurysm inflow angle. The association, if any, between αICA@PCOM and the aneurysm inflow angle requires additional research and will be the focus of future studies.

It has been previously shown that vessel morphology and local hemodynamics are closely related and play an important role in the initiation and growth of intracranial aneurysms.1 In particular the interplay between WSS and WSSG has been shown to induce endothelial cellular responses which can contribute to vascular pathology.25 26 CFD analysis on both parametric and patient-derived data show that increased vessel angulation or bending results in a pattern of high focal pressure, with low corresponding WSS and elevated adjacentWSSG, as evident in figures 2–3. The co-localization of high normalized pressure and low normalized WSS was previously correlated with thin-walled regions in cerebral aneurysms27 and may provide one explanation regarding the mechanism for changes in endothelial physiology leading to aneurysmogenesis.28 Healthy models displayed fewer hemodynamic variations at the location of the PCOM.

Local vessel morphology informs treatment options of intracranial aneurysms. Highly angular morphology was previously shown to influence the risk of incomplete stent apposition in Enterprise stent-assisted coiling and should be taken into account when selecting an endovascular strategy.29 In turn, stent implantation significantly decreases vessel curvature, remodeling the vessel and changing hemodynamic patterns, potentially contributing to favorable aneurysm management.20 30 These studies suggest the importance of understanding the patterns associated with ICA morphology at PCOM origin in the presence of PCOM aneurysms. Additional research will be needed to determine the clinical implications and applications of our findings.

Study limitations

As with similar retrospective research, the association between vessel angulation and aneurysm presence suggests a correlation, but does not demonstrate causation, despite our suggestive CFD results. There is the possibility that aneurysm presence modifies the morphology of the vessel and results in an increase bending of the ICA at the communicating segment. Our results showed this is unlikely as αICA@PCOM proved to be independent of aneurysm size. In addition, it should also be noted that the non-aneurysmal contralateral had significantly larger αICA@PCOM compared to healthy controls, which suggests that vessel bending precedes the aneurysm formation and contributes to hemodynamic conditions associated with aneurysm initiation.

Conclusion

The ICA morphology at the site of the PCOM is characterized by a significantly sharper bend (larger αICA@PCOM) in patients with PCOM aneurysms compared with contralateral and healthy ICAs. An optimal angle threshold of 61° (87% specificity, 80% sensitivity) was established in discriminating between aneurysmal samples and healthy controls. Ruptured aneurysms had significantly larger αICA@PCOM compared with unruptured aneurysms. Hemodynamically, increased bending of the ICA at the PCOM location results in a focally increased pressure and low WSS with sharp WSSG at the bend corresponding to the aneurysm neck. Future studies focusing on the prospective predictive performance of αICA@PCOM as a morphological biomarker will be needed to establish its clinical utility in informing clinical decision-making.

References

Footnotes

  • Contributors All authors are justifiably credited with authorship according to the authorship criteria. In detail: RR, acquisition of data, image processing, measurements; GC, acquisition of data, image processing, measurements; RM, image processing, measurement; JMB, acquisition of data; EL, acquisition of data; AL, analysis and interpretation of data, drafting of the manuscript, final approval given; AMM, conception, design, analysis and interpretation of data, critical revision, final approval.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are available upon reasonable request. Study data are from Tufts Medical Center clinical imaging repository, and as such contain patient identification, and cannot be shared in their raw state. De-identified models may be available upon reasonable request.