Comparison of an automatic analysis and a manual analysis of conjunctival microcirculation in a sheep model of haemorrhagic shock

Abstract

BackgroundLife-threatening diseases of critically ill patients are known to derange microcirculation. Automatic analysis of microcirculation would provide a bedside diagnostic tool for microcirculatory disorders and allow immediate therapeutic decisions based upon microcirculation analysis.MethodsAfter induction of general anaesthesia and instrumentation for haemodynamic monitoring, haemorrhagic shock was induced in ten female sheep by stepwise blood withdrawal of 3 × 10 mL per kilogram body weight. Before and after the induction of haemorrhagic shock, haemodynamic variables, samples for blood gas analysis, and videos of conjunctival microcirculation were obtained by incident dark field illumination microscopy. Microcirculatory videos were analysed (1) manually with AVA software version 3.2 by an experienced user and (2) automatically by AVA software version 4.2 for total vessel density (TVD), perfused vessel density (PVD) and proportion of perfused vessels (PPV). Correlation between the two analysis methods was examined by intraclass correlation coefficient and Bland-Altman analysis.ResultsThe induction of haemorrhagic shock decreased the mean arterial pressure (from 87 ± 11 to 40 ± 7 mmHg; p < 0.001); stroke volume index (from 38 ± 14 to 20 ± 5 ml·m−2; p = 0.001) and cardiac index (from 2.9 ± 0.9 to 1.8 ± 0.5 L·min−1·m−2; p < 0.001) and increased the heart rate (from 72 ± 9 to 87 ± 11 bpm; p < 0.001) and lactate concentration (from 0.9 ± 0.3 to 2.0 ± 0.6 mmol·L−1; p = 0.001). Manual analysis showed no change in TVD (17.8 ± 4.2 to 17.8 ± 3.8 mm*mm−2; p = 0.993), whereas PVD (from 15.6 ± 4.6 to 11.5 ± 6.5 mm*mm−2; p = 0.041) and PPV (from 85.9 ± 11.8 to 62.7 ± 29.6%; p = 0.017) decreased significantly. Automatic analysis was not able to identify these changes. Correlation analysis showed a poor correlation between the analysis methods and a wide spread of values in Bland-Altman analysis.ConclusionsAs characteristic changes in microcirculation during ovine haemorrhagic shock were not detected by automatic analysis and correlation between automatic and manual analyses (current gold standard) was poor, the use of the investigated software for automatic analysis of microcirculation cannot be recommended in its current version at least in the investigated model. Further improvements in automatic vessel detection are needed before its routine use.Electronic supplementary materialThe online version of this article (doi:10.1186/s40635-016-0110-5) contains supplementary material, which is available to authorized users.

A new complimentary web‐based tool for manual analysis of microcirculation videos: Validation of the Capillary Mapper against the current gold standard AVA 3.2

“…Nevertheless, analysis took twice as long with the Capillary Mapper 1.3 than with automatic CCTools, as recently reported . However, as long as automatic analysis is associated with the current significant lack of accuracy and the results show considerable bias, the Capillary Mapper represents a valid alternative to the gold standard AVA 3.2. Another advantage of the Capillary Mapper over current automatic analysis tools is its universal applicability.…”

“…The possible risk of a reduced quality of analysis, where microcirculation is compromised, for example, in septic or hemorrhagic shock, could therefore be excluded. In contrast, automatic microcirculation analysis software was recently shown to fail to discriminate between microcirculation in healthy animals and under hemorrhagic shock …”

“…However, the analysis of microcirculatory videos is still dependent on the individual investigator, because available programs allow only manual or semi‐manual analyses. This risk of bias would be eliminated with software for automatic analysis, but such programs have so far proven insufficiently accurate on testing . In addition, the few commercially available software packages that have been validated are quite expensive .…”

“…Users have therefore turned to standard image processing programs lacking specific validation . Because of its usability and reasonable analysis time, the commercial software AVA 3.2 (Microvision Medical B.V., Amsterdam, the Netherlands) is most commonly used for manual analysis of microcirculation videos and frequently taken as the gold standard …”

“…This risk of bias would be eliminated with software for automatic analysis, but such programs have so far proven insufficiently accurate on testing. 6,7 In addition, the few commercially available software packages that have been validated are quite expensive. 3,5,8 Users have therefore turned to standard image processing programs lacking specific validation.…”