While the computed tomography (CT) pulmonary manifestations of the Coronavirus disease 2019 (COVID-19) have been widely described (1–6), vascular manifestations have been less frequently reported and remain poorly understood. Multiple vascular changes are described in the literature, including mainly pulmonary embolism (7–10), vascular congestion or enlargement (11–17), and perfusion changes (18–21). Until now, the link between morphological changes and the correlation between biological changes and virus-induced inflammatory disorders remains an area of research. Furthermore, the COVID-19 acute respiratory distress syndrome (ARDS) is different from a typical pre-COVID ARDS, with relatively preserved lung volumes compared to the state of respiratory failure (22–24). Those findings suggest that a ventilation-perfusion mismatch phenomenon is not entirely responsible for the pathophysiology, but other vascular changes are implied in the process.

In this work, we investigate the blood volumes between healthy (COVID-19-free parenchyma) and diseased (COVID-19 alveolar opacity) lung regions in patients with COVID-19 pneumonia subjected to a computed tomography angiography (CTA). Our hypothesis is that the activation of arteriovenous anastomoses in affected lung regions under hypoxic conditions may translate into an augmented regional vascular volume on imaging.

Out of the 27 patients screened for inclusion, we excluded 17 due to excessive lung consolidation. We thus included 10 patients with images of predominant ground glass opacities. Patient’s characteristics are detailed in Table 1.

Computer-aided lung volume segmentation provided satisfactory results, with adequate delimitation between the lung and the non-lung structures. Total lung volumes (TLV) ranged from 1723 to 3,504 mL (mean: 3558 mL, ± 973 mL). Manually segmented total COVID volumes (TCV) ranged from 571 to 2,704 mL (mean: 1730 mL, ±748 mL). Subtracted total non-COVID volumes (TNCV) ranged from 451 to 2,849 mL (mean: 1829 mL, ±986 mL).

Vascular volume segmentation also provided adequate results with minimal overlap between lung parenchyma and vessels. Segmentation thresholds ranged from −10 to –150HU (mean: –65HU, ± 54HU). Extracted total lung vascular volumes (TVV) ranged from 71 to 388 mL (mean: 185 mL, ±93 mL). Extracted COVID vascular volumes (CVV) ranged from 49 to 314 mL (mean: 139 mL, ±82 mL). Subtracted healthy vascular volumes (NCVV) ranged from 3 to 79 mL (mean: 46 mL, ±27 mL).

The proportion of vascular volumes within diseased lung volumes (C_Vasc) ranged from 4 to 17% (mean: 8, ±5%), while the proportion of vascular volume within non-COVID lung volumes (NC_Vasc) ranged from 1 to 9% (mean: 3, ±2%) (Figure 2). C_Vasc was systematically higher than NC_Vasc, with differences (D_Vasc) ranging from 1 to 14% (mean: 5, ±4%) and ratios (R_Vasc) ranging from 1.3 to 5.95 (mean: 3.7, ± 2.08). All results are displayed in Table 2.

Statistical analysis using a Shapiro–Wilk test showed a normal data distribution, however due to small sample size, a non-parametric test was used. Wilcoxon rank-sum test showed significant differences between C_Vasc and NC_Vasc (p = 0.026).

In this study, we assessed the difference of vascular volumes between normal and affected lung parenchyma in COVID-19 patients. We found that affected zones showed significantly increased vascular volumes compared to normal zones. Vascular disorders in COVID-19 are still being investigated, and their distribution and prevalence are debated. Pulmonary embolism is one of the most researched topics related to vascular changes in COVID-19. In a multicenter study including 413 patients, Sadjad and al. found a pulmonary embolism incidence of 25% in hospitalized patients (7), while a systematic literature search identified in 27 studies with 3,342 patients a prevalence of 16.5% in all patients scanned for COVID-19 (9). Recently, Nevesny et al. (17) highlighted a potential connection between pulmonary embolism and venous congestion in the affected territories without arterial dilatation (17). More importantly, Nevesny et al. demonstrated that vascular congestion primarily happens in the venous compartment in COVID-19 pneumonia, with normal caliber veins in the non-affected zones, directing the discussion of pathophysiology towards specific mechanisms. The venous congestion they measured was at the segmental and subsegmental level, due to the difficulty to measure smaller vessels, which is something we also visually observed, but it could affect the whole venous tree and even the capillaries. Such venous congestion associated with paradoxical regional oligemia have raised the question of vascular shunting. Pre-existing, dysregulated arteriovenous anastomoses have been suggested as another phenomenon contributing to the vascular changes in COVID-19 (26–32). One theory is based on COVID-19 induced inflammation at the level of the capillaries, possibly inducing nitric oxide synthase (NOS) and nitric oxide (NO) production (33). This hypothesis describes increased NO activates precapillary arteriovenous anastomoses, usually present during fetal life and partially regressing after birth, leading to a right to left shunt and thus causing dead spaces and worsening hypoxemia (26, 30–32). This hypoxemic cascade explains the discrepancies between severe respiratory failures despite relatively moderate ventilation loss (22). More recently, Ackermann et al. demonstrated increased bronchopulmonary shunting, in other words arteriovenous anastomoses between bronchial arteries and pulmonary veins in hypoxic COVID-19 lung area, based on dual-energy CT, phase-contrast tomography, histology, and scanning electron micrography (34). Another hypothesis would be that venous obstruction or constriction would induce upstream microvascular congestion and elevated vascular pressure, leading to A-V shunts recruitment. While we did not observe this at the venous level, it could affect small veins, venules or venous capillaries, which we are not able to see with CT imaging.

Vascular changes in the lungs can be studied in many different ways. Our study focuses on the blood volume discrepancies between COVID and non-COVID parenchymal areas. We included the whole vascular tree in our vascular volumes, i.e., pulmonary arteries, arterioles, capillaries, venules, and pulmonary veins, as well as bronchial vascularization to a lesser extent. Thus, any change in vascular volumes between healthy and diseased volumes could be due to changes in any compartment. Other approaches, for example, the analysis of lung perfusion changes or the segmentation of only certain types of vessels with defined diameter thresholds, are possible.

In a study including 563 chest CTA, Poletti and al. tested an automated segmentation method and found an increased total blood volume and increased volume of larger lung vessels in COVID-19 pneumonia compared to non-COVID 19 patients (35). Another study by Muriel and al. showed increased blood volumes in small vessels between 5 and 10 mm² and above 10 mm² (36).

In a retrospective study including 48 patients, Lang and al. found dilated distal vessels in 41 patients, with a mix of regional hyperemia in the affected zones in 13 patients and oligemia in 24 patients (37).

In a pilot study using Dual Energy CT (DECT), Si-Mohamed and al. observed increased perfusion in the affected parenchyma during the early phase of COVID-19 and decreased perfusion after 2 weeks (21).

Our study shows similar results. Our major findings are the significant difference in vascular volumes proportions between COVID (C_vasc) and non-COVID (NC_vasc) areas. As can be seen in Table 2 and Figure 2, C_vasc was systematically higher in the affected regions with a mean difference (D_vasc) of 5% and a mean ratio (R_vasc) of 3.7. These results can also be appreciated in the segmented volume images, such as in Figure 1 where venous congestion with increased vascular volumes are seen COVID affected areas. Such findings support the hypothesis of the presence of recruited arteriovenous shunts due to the local inflammatory state and/or local hypoxia of the lung parenchyma.

Coexistence of hypoxemia, pulmonary venous hypertension, endothelial inflammation/injury and predominantly pulmonary venous thrombosis might suggest a sequence involving selective active vasoconstriction of small-to intermediate-size pulmonary veins leading to venous microvascular congestion, blood flow stasis and thrombosis as well as pulmonary venous hypertension and hypoxemia that is exaggerated by opening arteriovenous anastomoses.

Although we put an emphasis on recruitment of pulmonary arteriovenous anastomoses, multiple other processes could co-exist and participate in the difference of vascular volumes. For example, vasoactive mediators such as mast-cells derived cytokins, angiotensin II (Ang II), vasoactive intestinal peptide (VIP), endothelin-1 (ET-1), etc. could be involved in this process (38–41). Thromboxane A2 and F2-isoprostanes are two other important vasomediators known to induce venoconstriction via activation of TP receptors in COVID-19 patients (42–44), It is relevant that the TP receptor blocker, ramatroban, was found in anecdotal clinical experience to relieve respiratory distress and hypoxemia in COVID-19 patients (45).

Other factors could also participate or potentialize the recruitment of vascular anastomoses, such as microthrombosis, endotheliitis or a mix of increased increased pulmonary pressure and cardiac output (32, 46, 47).

Our study is subject to several limitations. First, the number of included patients is low as we only included 10 patients; however, this was sufficient, given the statistically significant differences we found. As a consequence, we did not include additional cases. Second, the study’s retrospective nature is a potential source of bias. Third, we excluded patients presenting with dense consolidation in COVID affected regions as it resulted in over-segmentation, with an overlap that would probably result in an overestimation of the vascular volumes discrepancies between COVID and non-COVID territories. Another method should be used if one wants to assess vascular volumes in patients with more consolidation (i.e., patients in a later course of disease), which could result in interesting data for another study. Fourth, we segmented the whole vascular tree and could thus not determine whether the observed differences were more prevalent in the vessels’ capillary, venule or venous sections. Finally, segmentation thresholds were defined manually, and while we kept the same value in each patient’s left and right lung, there was interpatient variation. An artificial intelligence based automatic segmentation could provide better results, but such algorithm does not currently exist. While we focused on CT modality for this paper, further works could explore other modalities to confirm our findings, such as MRI or doppler echography.

Vascular volume, obtained by semi-automatic segmentation in COVID-19 pneumonia was significantly higher in areas affected by alveolar opacification than lung segments appearing normal. We found a 3.7 mean ratio of proportion of vascular volume between COVID and non-COVID affected areas. These results are consistent with other reports mentioning venous enlargement, increased lung perfusion in affected zones, and the supposed recruitment of pre-existing intrapulmonary arteriovenous shunts that could explain the discrepancies between the morphological disease severity on imaging and the clinical presentation of the patients.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethics Committee of Vaud, Switzerland (project-ID 2020 -01469, 24 November 2020). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

DR and SQ: conceptualization and supervision. DR: methodology. SQ, GF, and DR: validation. A-CR and LG: investigation. GF: writing—original draft preparation. DR and CP: writing—review and editing. All authors contributed to the article and approved the submitted version.

Open access funding by University of Lausanne.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.