INTRODUCTION

Throughout the COVID-19 pandemic, a portion of those affected have evolved toward acute hypoxic respiratory failure (1). Initially, this was hypothesized to result from acute lung injury leading to acute respiratory distress syndrome (ARDS), with many patients meeting the imaging requirements of the Berlin criteria for ARDS (2, 3). Treatment protocols for these patients were therefore designed based on the current guidelines for ARDS management, including mechanical ventilation with high levels of positive end-expiratory pressure (PEEP) (3). A growing number of reports suggest that some patients with COVID-19 present with relatively normal lung compliance but severe and refractory hypoxemia, which is inconsistent with the conventional understanding of ARDS, indicating that an alternate pathophysiology may be involved (4–6).

In previous research, a novel quantitative CT postprocessing technique was described to quantify the volume of blood contained within pulmonary blood vessels of a given size (7). In comparing scans of 103 patients with PCR-confirmed COVID-19 to 108 healthy volunteers, it was observed that patients with COVID-19 have significantly less blood (as a portion of total pulmonary blood volume) contained in pulmonary vessels below 5 mm² in cross-sectional area (“BV5”), and significantly more in vessels greater than 10 mm² in cross-sectional area (“BV10”) (Fig. 1A). These changes are equivalent to a redistribution of pulmonary blood away from the small pulmonary vessels and into larger vessels, consistent with increased pulmonary vascular resistance downstream. This may be due to dysregulated vasoconstriction, thrombotic events in the microvasculature, or both. The endothelial inflammatory pathways may play an important role in this dysregulation (8). Any of these could contribute to the observed refractory hypoxemia by impairing gas exchange across the alveolar-capillary membrane. These data suggest a possible pathophysiological explanation for the impression of severe dead space ventilation reported by intensivists (6).

Considering these previous data, the relationship (or predictive value) between CT-derived measures of small blood vessel loss and the degree of gas exchange anomaly experienced was investigated. Paired clinical data were analyzed from 43 of the 103 patients previously analyzed (7), as well as eight additional patients included after the previous research concluded. We hypothesized that patients with lower BV5 blood flow would have higher supplemental oxygen needs and less favorable arterial blood gas profiles.

METHODS

From the initial data analysis, 111 hospitalized COVID-19 patients were retrospectively selected based on the availability of CT scans of the lungs with slice thickness of 1.5 mm or less, as well as PCR-confirmed SARS-CoV2 infection. Thin-sliced CT scans were provided by the respective hospitals where they were acquired. Institutional review board approval was granted by the respective local committees. Written informed consent was obtained from the patients, volunteers, and/or institutions included in the article. Ninety-one scans came from Belgium [51 from AZ Sint-Maarten (Mechelen), 40 from Ziekenhuis Oost-Limburg (Genk)], 10 from the United Kingdom (Royal Papworth Hospital), and 10 from China (Wenzhou Medical University). It should be noted that the previous work included only 43 patients from AZ Sint-Maarten; the additional eight were provided after that research concluded. One hundred eight inspiratory scans from healthy patients were acquired from the COPDGene cohort (NCT00608764) and were used as reference data. No statistical comparison between these data and COVID-19 data set was performed (9, 10). Because scans were acquired without a standardized protocol, slice thickness varied between 0.6 mm and 1.5 mm. One patient had highly anomalous arterial blood gas readings (assumed to be a mixed venous sample) and was therefore excluded from the analysis. The measurements were performed on the TLC level without gaiting as this was not possible in this clinical setting. All patients, however, were instructed and encouraged during the scanning procedure to inhale to their maximal inspiration.

3-D reconstructions of the lungs and pulmonary vasculature were created (Fig. 2). An automated blood vessel segmentation algorithm performed an eigenvalue analysis of the Hessian matrix to enhance and identify tubular structures, by returning the probability of each voxel belonging to tubular structure based on shape analysis (11). Hounsfield unit (HU) thresholds, based on the vessels size defined by an automated adaptive iterative threshold method, were used to limit the vessels. During preprocessing, a gradient anisotropic diffusion filter was applied, and a region of interest was defined to remove false positives. Subsequently, the smaller nonconnected parts were removed. To account for the effects of slice thickness on results, sensitivity analysis was performed. Volumes were computed from the cross-sectional area of each vessel. Denoting these measurements “BVX”, where “X” indicates a range of vessel sizes in mm² (BV5 is the volume of blood contained in vessels between 1.25 and 5 mm² cross-sectional area, BV5-10 between 5 and 10 mm², and BV10 > 10 mm²).

To account for variation in lung volume, BVX was normalized to total pulmonary blood volume, allowing for the computation of a “BV spectrum,” a curve representing the percent of total pulmonary blood volume contained within vessels of a given caliber as a function of cross-sectional area. This method was previously described and was correlated to the clinical symptoms of patients as validation of the technique.

Further analysis was performed on 50 patients (30 male, 20 female, median age 61.98) included from AZ Sint-Maarten, as these were the patients for whom the most clinical data was available. These patients were divided into groups based on their need for oxygen at the time of CT scan acquisition, with a threshold of 2 L/min chosen a priori (<2 L/min O₂ or >2 L/min O₂). This threshold was chosen as patients in need of more than 2 L/min are oxygen dependent. Those below 2 L/min have a more intermittent need for oxygen. The oxygen needs of patients with more than 2 L vary over time. We chose to pool patients in this study, rather than analyze the amount of oxygen needed as a continuous variable. BV5 values were compared between cohorts. Nonparametrical spearman correlations were computed between BV5 and $\text{P}{\text{a}}_{{\text{O}}_2}$ AaDO2. Four patients did not have arterial blood gas data available and were therefore excluded from this portion of the analysis. Two-sample t tests were used to assess significance (P < 0.05). All analyses were performed using the open-source statistical environment R v. 3.2.5 or higher (The R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Eighteen out of 50 patients needed >2 L/min supplemental oxygen and this group demonstrated a significantly lower median percentage of total blood flow in the BV5 vessels compared with the 32 patients who needed <2 L/min supplemental oxygen (41.61% vs. 46.89%, P = 0.023). Both groups had significantly less blood as a proportion in BV5 vessels compared with healthy volunteers (Fig. 1B). The 46 patients for whom arterial blood gas data were available showed that decreased percentage of blood distribution through BV5 vessels correlated with both increased alveolar-arterial oxygen gradient (AaDO2) (Fig. 1C) and decreased $\text{P}{\text{a}}_{{\text{O}}_2}$ (Fig. 1D).

DISCLOSURES

W.D., Y.D.M., J.D.B., B.L., and M.Lanclus are employees of FLUIDDA, a company that develops and markets part of the technology described in this article. The other authors have no financial relationships with any organization or company that might have an interest in the submitted work and received no direct funding from FLUIDDA. None of the other authors has any conflicts of interest, financial or otherwise, to disclose.

AUTHOR CONTRIBUTIONS

W.D., W.D.B., M.Lins, J.D.B., E.F., and B.R.L. conceived and designed research; W.D., M.Lins, K.I., J.V., J.D.B., E.F., B.R.L., M.Lanclus, and M.T. performed experiments; W.D., M.Lins, K.I., J.D.B., E.F., B.R.L., M.Lanclus, and M.T. analyzed data; W.D., W.D.B, M.Lins, K.I., J.D.B., E.F., B.R.L., M.Lanclus, and M.T. interpreted results of experiments; W.D., W.D.B., J.D.B., and B.R.L. prepared figures; W.D., W.D.B., J.V., J.D.B., and B.R.L. drafted manuscript; W.D., W.D.B., Y.D.M., K.I., J.V., J.D.B., E.F., B.R.L., M.Lanclus, and M.T. edited and revised manuscript; W.D., W.D.B., M.Lins, Y.D.M, K.I., J.D.B., E.F., B.R.L., M.Lanclus, and M.T. approved final version of manuscript.