Preliminary clinical evaluation of automated analysis of the sublingual microcirculation in the assessment of patients with septic shock: Comparison of automated versus semi-automated software

Abstract

INTRODUCTION:

The outcome of patients in septic shock has been shown to be related to changes within the microcirculation. Modern imaging technologies are available to generate high resolution video recordings of the microcirculation in humans. However, evaluation of the microcirculation is not yet implemented in the routine clinical monitoring of critically ill patients. This is mainly due to large amount of time and user interaction required by the current video analysis software. The aim of this study was to validate a newly developed automated method (CCTools^®) for microcirculatory analysis of sublingual capillary perfusion in septic patients in comparison to standard semi-automated software (AVA3^®).

METHODS:

204 videos from 47 patients were recorded using incident dark field (IDF) imaging. Total vessel density (TVD), proportion of perfused vessels (PPV), perfused vessel density (PVD), microvascular flow index (MFI) and heterogeneity index (HI) were measured using AVA3^® and CCTools^®.

RESULTS:

Significant differences between the numeric results obtained by the two different software packages were observed. The values for TVD, PVD and MFI were statistically related though.

CONCLUSION:

The automated software technique successes to show septic shock induced microcirculation alterations in near real time. However, we found wide degrees of agreement between AVA3^® and CCTools^® values due to several technical factors that should be considered in the future studies.

Keywords

Sublingual microcirculation sepsis shock capillary perfusion incident dark field imaging

1 Introduction

The prognostic value of studies of the microcirculation in critically ill patients besides other parameters measured at the bedside such as systemic hemodynamics or oxygen derived parameters has been recently identified. Furthermore, several clinical studies have shown the impact of standard as well as innovative sepsis therapies on the parameters of the sublingual microcirculation [1 –8].

More than 500 experimental and clinical microcirculatory studies were published in the last 15 years. However, evaluation of the sublingual microcirculation is still not implemented yet in the routine clinical monitoring of critically ill patients. Side-stream dark field (SDF) imaging represents the most widely used method in the past and the analysis of the video recordings with the provided semi-automated AVA3^® software counts as the “gold standard”. The obtained video recordings of the blood flow in sublingual microvessels have been validated in identifying the severity of sepsis-related disturbances of the microcirculation [9, 10]. However, semi-automated analysis requires a large amount of user interaction, which increases observer bias and analysis time.

Recently, Aykut and co-workers [11] validated a novel lightweight handheld microscope device which uses incident dark field (IDF) illumination and automated analysis software (CCTools^®). Their results showed that IDF imaging was able to detect more capillaries in the sublingual microcirculation (capillary density) due to improved image quality. In the meantime, automated analysis software is also available for SDF imaging. Only with automated software changes in tissue perfusion can be monitored in real-time and can be used as a diagnostic and a therapeutic endpoint in clinical routine. However, the newly developed automated analysis methods for sublingual capillary perfusion have not been evaluated in septic patients yet.

The aim of this study was to validate the most recent imaging technology, i.e. IDF, and the provided automated software (CCTools^®) in patients with sepsis by comparing the main key microcirculatory parameters to corresponding values obtained by semi-automated software.

2 Patients and methods

The study protocol was reviewed and approved by both Canadian REB (Dalhousie University, Halifax, NS, 2013-3161) and the local Institution Ethical Committee (Cairo, Egypt, N-16-2014). Informed consent had been obtained from the patient’s relatives.

2.1 Study design and setting

Single-center, observational study conducted November 2014 – May 2015. Surgical Intensive Care Trauma Center (TS-ICU-185), Teaching Hospital, Faculty of Medicine, Cairo University, Egypt.

2.2 Selection of participants

Forty seven patients older than 18 years meeting the criteria of septic shock according to the Surviving Sepsis Campaign [12].

Exclusion criteria were liver cirrhosis, shock due to other causes, acute ischemic event present at the time of diagnosis, and oral injuries. Patients were enrolled consecutively depending on the capability of the attending intensivist to perform IDF imaging once patients were stabilized. All patients were mechanically ventilated and sedated.

2.3 IDF imaging device

Sublingual microvascular video recordings were obtained using the Cytocam-IDF imaging device (Braedius Medical, Huizen, The Netherlands). IDF uses incident dark field illumination with high-brightness light emitting diodes (LED’s) as a source of light, and with a very short illumination pulse time of 2 milliseconds. The Cytocam-IDF has a 4X magnification objective. Image acquisition is computer controlled and electronically synchronized to the illumination pulses. This feature, and a specialized set of lenses, result in high penetration sharp contour visualization of the blood cells within the microcirculation. The optical system provides an optical resolution of more than 300 lines/mm. The IDF probe is a handheld device. Optimal focus depth and brightness are default adjusted (160 μm, 450 respectively). Video sequences of 4 seconds, from a different sublingual sites, were acquired and saved for offline analysis. The Cytocam-IDF imaging device is supplied with an analysis software for automated acquisition and quantification of microcirculatory parameters (CCTools^®, see below).

2.4 Video acquisition, processing and analysis

Key points for optimal quality of the video recordings – as defined in the consensus round table report [10] – are avoidance of pressure artefacts and secretions, adequate focus and contrast adjustment, and frame size reduction. Video sequences with low quality were discarded.

All videos were recorded by five previously trained persons (A.M., R.M., H.M., M.A., A.A.). All video files were given generic names [ICU-Sepsis-Alphabetical name- Time] by name generator, and generated names only were used in further evaluation.

The video clips were also exported to Audio Video Interleave (AVI) format (resolution 750×576) for blind analysis by expert investigators (S.I., C.L.) using AVA3^® software (MicroVision Medical; Amsterdam; The Netherland).

Total vessel density (TVD) was estimated as the total length of vessels divided by the total surface of the area, perfused vessel density (PVD) were determined as the total vessel density multiplied by the fraction of perfused vessels. The proportion of perfused vessels (PPV [% ]) was calculated as follows: 100 × (total number of vessels – [no flow + intermittent flow])/total number of vessels. Perfusion is categorized – according to type of flow in individual vessel - as no flow (0), intermittent flow (1), sluggish flow (2), and continuous flow (3). The heterogeneity index was calculated as the highest site flow velocity minus the lowest site flow velocity, divided by the mean flow velocity of all sublingual sites [10].

2.5 Statistical analysis

Statistical analyses were performed using GraphPad Prism statistical software version 5.0 for Windows (GraphPad Inc, San Diego, CA, USA). The Kolmogorov-Smirnov test was used to confirm normal distribution. Student’s t-test was used for comparisons between groups. Furthermore, degree of agreement, correlation, linear regression, and Bland– Altman analyses were performed. Data were presented as the mean and standard error. A p value < 0.05 was considered statistically significant.

3 Results

3.1 Demographics and clinical measurements

Our population baseline demographic characteristics, hemodynamic and metabolic variables are summarized in Tables 1, 2.

Table 1
Characteristics of the patients

Septic shock patients (47)

Male n (%) 29 (60)

Age (yrs) 50.5 [40.5, 62.5]

BMI 27.7 [25.9, 32.7]

Co-morbidity, n (%)

DM 14 (29)

HTN 5 (10)

others 5 (10)

Sedation/analgesia n (%) 47 (100)

Mechanical ventilation, n (%) 47 (100)

Temperature (°C) 37.0 [37.0, 37.5]

Infection Source n (%)

Lung 3 (6)

Abdomen 34 (72)

Urinary –

others 10 (21)

APACHE 23 [17, 27]

SAPS 49 [32, 65]

	Septic shock patients (47)
Male n (%)	29 (60)
Age (yrs)	50.5 [40.5, 62.5]
BMI	27.7 [25.9, 32.7]
Co-morbidity, n (%)
DM	14 (29)
HTN	5 (10)
others	5 (10)
Sedation/analgesia n (%)	47 (100)
Mechanical ventilation, n (%)	47 (100)
Temperature (°C)	37.0 [37.0, 37.5]
Infection Source n (%)
Lung	3 (6)
Abdomen	34 (72)
Urinary	–
others	10 (21)
APACHE	23 [17, 27]
SAPS	49 [32, 65]

All values are expressed in median and inter-quartile range (IQR). BMI; body mass index, DM; diabetes mellitus, HTN; hypertension, APACHE; acute physiology and chronic health evaluation, SAPS; simplified acute physiology score.

Table 2

Hemodynamic and metabolic data of the patients

	Septic shock patients (47)
HR (bpm)	106 [90, 128]
MBP (mmHg)	80.5 [88.5, 105]
CVP (mmHg)	13 [10, 15]
Vasopressor use, n (%)	47 (100)
Vasopressor dose, (mcg/kg/min)	0.3 [0.2, 0.7]
Hbg (g/dL)	10.3±2
Hct (%)	31±6.4
WBC (10⁹/L)	21.5±20.2
Arterial O₂ sat (%)	98 [96.2, 99]
PaO₂ (mmHg)	119 [85.2, 155.8]
Scvo₂%	68 [58,75.7]
PaCO₂ (mmHg)	31 [25.5,37]
pH	7.2 [7.2,7.4]
Lactate (mmol/l)	4 [2, 6]

All values are expressed in median and inter-quartile range (IQR) and in mean±SD. HR; heart rate, MBP; mean arterial blood pressure, CVP; central venous pressure, Hbg; hemoglobin, Hct; hematocrit, WBC; white blood cell count, PaO₂; partial pressure of oxygen in arterial blood, Scvo_2 ; central venous oxygen saturation, PaCO₂; partial pressure of carbon dioxide in arterial blood.

Forty seven patients were enrolled (29 (60%) male; age 50.5 [40.5, 62.5]; BMI 27.7 [25.9, 32.7]; 49% of patient had history of chronic disease). Main source of sepsis was abdominal infection – 72%; APACHE 23 [17, 27] and SAPSE 49 [32, 65]. Among the systemic hemodynamic and perfusion parameters, tachycardia (HR, 106 [90, 128]) bpm, metabolic acidosis and hyperlactatemia (lactate, 4 [2, 6] mmol/L) were observed in the study patients. NE requirements were 0.3 [0.2, 0.7] μg/kg/minute.

3.2 Microcirculatory parameters

The obtained vascular density parameters, TVD and PVD, were significantly different depending on the software used for analysis (TVD (mm/mm²) - AVA3:11.3±0.2; CCTools 10.1±0.2); (PVD (mm/mm²) - AVA3:10.6±0.2; CCTools 8.0±0.2; Table 3).

Table 3
Comparison of the microcirculatory parameters obtained by semi-automated software (AVA3) vs automated method (CCTools)

AVA3 CCTools

TVD [mm/mm²] 11.3±0.2 10.1±0.2^#

PVD [mm/mm²] 10.6±0.2 8.0±0.2^#

PPV (%) 92.18±1.6 77.2±1.3^#

MFI 2.6±0.04 2.0±0.04^#

Heterogeneity index 0.7±0.07 0.8±0.3

	AVA3	CCTools
TVD [mm/mm²]	11.3±0.2	10.1±0.2^#
PVD [mm/mm²]	10.6±0.2	8.0±0.2^#
PPV (%)	92.18±1.6	77.2±1.3^#
MFI	2.6±0.04	2.0±0.04^#
Heterogeneity index	0.7±0.07	0.8±0.3

All values are expressed in mean and standard error. TVD; total vessel density, PVD; perfused vessel density, PPV; proportion of perfused vessel, MFI; microvascular flow index. ^#<0.05 vs. AVA3.

Moreover, perfusion parameters showed significant differences between the two methods (PPV (%) - AVA3: 92.18±1.6; CCTool 77.2±1.3; MFI - AVA3: 2.6±0.04; CCTool: 2.0±0.04; Table 3).

No significant difference was observed between AVA3 and CCTool calculated HI.

3.3 Correlation and agreement results

204 paired data points were included for statistical analysis. The Pearson correlation for the individual data points showed that microcirculatory parameters obtained from CCTools software correlated with their corresponding parameters reported by AVA3 programme (TVD: r = 0.3, R² = 0.1, P < 0.0001, 95% confidence interval (CI) for r of 0.2 to 0.4; PVD: r = 0.3, R² = 0.1, P < 0.0001, 95% confidence interval (CI) for r of 0.1 to 0.4; MFI: r = 0.2, R² = 0.08, P < 0.001, 95% confidence interval (CI) for r of 0.08 to 0.3]. PPV showed no significant correlation between CCTool and AVA3 [PPV: r = 0.1, R² = 0.04, P < 0.058, 95% confidence interval (CI) for r of – 0.0 to 0.26]. Moreover, we observed no significant correlation between CCTool and AVA3 for HI calculated values [HI: r = –0.1, R² = 0.01, P = 0.4, 95% confidence interval (CI) for r of –0.3 to 0.1]; Fig. 1).

Except for PPV, Bland-Altman plots of manually edited AVA3 vs. fully automated CCTool analysis showed that CCTool was capable of producing results comparable with those achieved from AVA3 (TVD (mm/mm²) - AVA3 vs CCTool: 1.2 [95% Limits of Agreement from –5.9 to 8.4]; PVD (mm/mm²) - AVA3 vs CCTool: 3.0 [95% Limits of Agreement from –11.9 to 18.0]; PPV(%) - AVA3 vs CCTool: 14.9 [95% Limits of Agreement from –37.4 to 67.2]; MFI - AVA3 vs CCTool: 0.5 [95% Limits of Agreement from –0.8 to 2.0]; HI - AVA3 vs CCTool: – 0.09 [95% Limits of Agreement from –4.8 to 4.6]); Fig. 2).

Fig.1

(A-E) Correlation of microcirculatory parameters analysed by AVA3 and CCTools: Density parameters (A, Total vessel density (TVD) and B, Perfused vessel density (PVD)), perfusion parameters (C, Proportion perfused vessel (PPV) and D, Mean Flow index (MFI)) and E, Heterogeneity Index (HI).

Fig.2

(A-E) Bland-Altman plots of microcirculatory parameters analysed by AVA3 vs. CCTools: Density parameters (A, Total vessel density (TVD) and B, Perfused vessel density (PVD)), perfusion parameters (C, Proportion perfused vessel (PPV) and D, Mean Flow index (MFI)) and E, Heterogeneity Index (HI).

3.4 Real-time assessment and video size estimation

Microcirculation parameters were measured in near real time using the CytoCam IDF imaging system (average: 5 minutes). The size of CCTools generated video files of 101 frames is 350 MB and the calculated storage size for individual patient is 1050 MB. Meanwhile, the size of exported AVI clip is 44.2 MB and calculated storage size for individual patient (3 recorded videos) is 132.6 MB.

4 Discussion

Our study - for the first time - presents an evaluation of a recently developed automated software method to study the sublingual microcirculation in septic patients. The automated software package, CCTools^® was used to analyze 204 video sequences and the obtained results were compared with the semi-automated analysis method, AVA3^®. We observed that the microcirculation parameters produced by the automated software are consistent to previous reported semi-quantitative descriptive results [13 –15]. However, significant differences between the parameters that were measured by the two investigated software packages were found.

The microcirculatory measurements in septic shock patients showed that both, the density (TVD and PVD) and MFI parameters, obtained by AVA3 or CCTools, respectively, are correlated. However, the TVD and PVD bias for CCTools were 1.2 and 3 respectively, and both have wide limits of agreement. CCTools seem to overestimate and sometime underestimate density parameters. The numeric differences in density measurements between AVA3 and CCTools can be related to the differences in the optical field of view between the original IDF image and the exported AVI images for AVA3 analysis.

Although the algorithm of CCTools demonstrates promises in its ability to rapidly provide quantitative information about the microcirculation (vessel diameter, length and density) at the bedside, there is still obvious incidence of false positive and false negative vessels detected by the software. The ability of the investigator to exclude vessels out of focus is an important feature of CCTools. Both CCTools and AVA3 include image stabilization and video editing properties. Image stabilization can resize the image so that the final image has a magnification different from that of the original image.

We observed also that MFI measurement points obtained from CCTools are scattered along the regression line. However, MFI bias for CCTool is 0.5, and the limit of agreement is narrow. We observed significant differences between CCTools and AVA3 measurements of MFI. The difference between CCTools measured and AVA3 estimated MFI could be due to individual bias and the estimation method. CCTools provides an automated preliminary MFI parameter, which is different from the classical MFI; it reflects the average speed index values for the full field of view area that previously described. Moreover, the classic classification of flow results in a non-linear scale. The classic MFI score is ordinal and discontinuous; it ranges from 0 to 3, and a change from 0 to 1 may not have the same implications for tissue perfusion as a change from 2 to 3, which may complicate the interpretation of the effects of therapeutic interventions. Meanwhile the preliminary CCTool MFI is based on a linear scale between 1 and 10 related to real flow behaviour measured as speed (micrometers per second).

The heterogeneous microcirculation is described in many disease states, especially in severe sepsis and septic shock. Shunt fraction is reflected by blood flow heterogeneity in the investigated area by PPV and between the different areas of the investigated organ by the heterogeneity index [16]. Trzeciak and coworkers [17] proposed the heterogeneity index (HI), which involves evaluating three to five sites and measuring the MFI in the quadrants, taking the difference between highest MFI minus the lowest site MFI divided by the mean flow velocity of all sublingual sites at a single time point. HI is a key determinant of the shunted fraction, often seen in distributive shock.

We found no relations between PPV and HI estimated by both CCTools and AVA3. Wide limit of agreement was noted with PPV, CCTools seems to overestimate and sometime underestimate PPV value. Meanwhile, narrow limit of agreement was noted with HI.

Investigators needed 1 to 5 minutes to get the average results of at least three videos for individual patient. Additionally, automated software does not require the considerable training that must be provided to the AVA3 user to be able to properly perform video analyses. The need for adequate harddrive storage capacity should be anticipated for CCTools videos.

The results of our observational study have some limitations: First, we did not use the original SDF technology for the AVA3 analyses. However, we observed that pressure artefacts are reduced and video quality is improved with IDF technology, an observation that is certainly related to the light weight Cytocam imaging probe in comparison to the much heavier SDF device. Our observation is consistent to previous studies [11, 18] comparing IDF and SDF technology. However, with our observational results we are not able to discuss hardware–related software algorithms. Our aim was to investigate clinical aspects of analyzing microcirculatory images of septic shock patients obtained by two different software packages. Sorelli et al. [19] had recently compared IDF -Kalman filter object tracking technique against space-time diagrams. They reported high correlation (r = 0.96) and agreement (bias = 3.3 μm/s).

5 Conclusion

In the present study, by using IDF technology we were able to adequately detect microcirculatory alterations in patients with septic shock in drastically reduced analysis time and observer bias. The real-time bedside quantitative evaluation of septic shock-induced microcirculation alterations by automated CCTools software had wide degrees of agreement with semi-automated AVA3 technique. The differences are related to several technical factors. Our study is an essential step to evaluate recent developed automated software in comparison to semi-automated software in septic shock patients. Further developmental works is needed to improve the automated software.

Authors’ contributions

NS and CL carried out the study design and performed the statistical analysis.

AM, RM, HM, MA, AH, OH, AR, AS and AH carried out patients recruitment and data extraction. NS, SI and CL participated in videos qualifications and analysis. SW participated in study coordination. NS, VC, RG and CL helped to draft the manuscript.

All authors read and approved the final manuscript.

Conflicts of interest

The author(s) declare that they have no competing interests.

Footnotes

Acknowledgments

This study was funded by Stars in Global Health; Grand Challenges Canada.

References

Dubois

, De Backer

, Creteur

, Anane

, Vincent

. Effect of vasopressin on sublingual microcirculation in a patient with distributive shock. Intensive Care Med. 2003;29:1020–3.

Ruiz

, Bruhn

, Hernandez

, Andresen

. Evaluation of sublingual microcirculation in septic shock. Report of one patient treated with high volume hemofiltration. Rev Med Chil. 2008;136:1175–8.

Maier

, Hasibeder

, Hengl

, Pajk

, Schwarz

, Margreiter

, Ulmer

, Engl

, Knotzer

. Effects of phenylephrine on the sublingual microcirculation during cardiopulmonary bypass. Br J Anaesth. 2009;102:485–91.

Morelli

, Donati

, Ertmer

, Rehberg

, Kampmeier

, Orecchioni

, Di Russo

, D’Egidio

, Landoni

, Lombrano

, Botticelli

, Valentini

, Zangrillo

, Pietropaoli

, Westphal

. Effects of vasopressinergic receptor agonists on sublingual microcirculation in norepinephrine-dependent septic shock. Crit Care. 2011;15:R217.

Atasever

, Boer

, Speekenbrink

, Seyffert

, Goedhart

, de Mol

, Ince

. Cardiac displacement during off-pump coronary artery bypass grafting surgery: Effect on sublingual microcirculation and cerebral oxygenation. Interact Cardiovasc Thorac Surg. 2011;13:573–7.

Forti

, Comin

, Lazzarotto

, Battistella

, Salandin

, Sorbara

. Pump flow changes do not impair sublingual microcirculation during cardiopulmonary bypass. J Cardiothorac Vasc Anesth. 2012;26:785–90.

Qian

, Yang

, Cahoon

, Xu

, Zhu

, Yang

, Hu

, Sun

, Tang

. Post-resuscitation intestinal microcirculation: Its relationship with sublingual microcirculation and the severity of post-resuscitation syndrome. Resuscitation. 2014;85:833–9.

Edul

, Ince

, Navarro

, Previgliano

, Risso-Vazquez

, Rubatto

, Dubin

. Dissociation between sublingual and gut microcirculation in the response to a fluid challenge in postoperative patients with abdominal sepsis. Ann Intensive Care. 2014;4:39.

Boerma

, Mathura

, van der Voort

, Spronk

, Ince

. Quantifying bedside-derived imaging of microcirculatory abnormalities in septic patients: A prospective validation study. Crit Care. 2005;9:R601–6.

10.

De Backer

, Hollenberg

, Boerma

, Goedhart

, Buchele

, Ospina-Tascon

, Dobbe

, Ince

. How to evaluate the microcirculation: Report of a round table conference. Crit Care. 2007;11:R101.

11.

Aykut

, Veenstra

, Scorcella

, Ince

, Boerma

. Cytocam-IDF (incident dark field illumination) imaging for bedside monitoring of the microcirculation. Intensive Care Med Exp. 2015;3:40.

12.

Dellinger

, Levy

, Rhodes

, Annane

, Gerlach

, Opal

, Sevransky

, Sprung

, Douglas

, Jaeschke

, Osborn

, Nunnally

, Townsend

, Reinhart

, Kleinpell

, Angus

, Deutschman

, Machado

, Rubenfeld

, Webb

, Beale

, Vincent

, Moreno

. Surviving Sepsis Campaign: International guidelines for management of severe sepsis and septic shock. Intensive Care Med. 2012;39:165–228.

13.

Sakr

, Dubois

, De Backer

, Creteur

, Vincent

. Persistent microcirculatory alterations are associated with organ failure and death in patients with septic shock. Crit Care Med. 2004;32:1825–31.

14.

Creteur

, De Backer

, Sakr

, Koch

, Vincent

. Sublingual capnometry tracks microcirculatory changes in septic patients. Intensive Care Med. 2006;32:516–23.

15.

Hernandez

, Boerma

, Dubin

, Bruhn

, Koopmans

, Edul

, Ruiz

, Castro

, Pozo

, Pedreros

, Veas

, Fuentealba

, Kattan

, Rovegno

, Ince

. Severe abnormalities in microvascular perfused vessel density are associated to organ dysfunctions and mortality and can be predicted by hyperlactatemia and norepinephrine requirements in septic shock patients. J Crit Care. 2013;28:538, e9–14.

16.

Spronk

, Zandstra

, Ince

. Bench-to-bedside review: Sepsis is a disease of the microcirculation. Crit Care. 2004;8:462–8.

17.

Trzeciak

, Dellinger

, Parrillo

, Guglielmi

, Bajaj

, Abate

, Arnold

, Colilla

, Zanotti

, Hollenberg

. Early microcirculatory perfusion derangements in patients with severe sepsis and septic shock: Relationship to hemodynamics, oxygen transport, and survival. Ann Emerg Med. 2007;49:88–98, e1–2..

18.

Hutchings

, Watts

, Kirkman

. The Cytocam video microscope. A new method for visualising the microcirculation using Incident Dark Field technology. Clin Hemorheol Microcirc. 2016;62:261–71.

19.

Sorelli

, Bocchi

, Ince

. Monitoring the microcirculation at the bedside using hand-held imaging microscopes: Automatic tracking of erythrocytes. Conf Proc IEEE Eng Med Biol Soc. 2015, 7378–81.