Skin color independent robust assessment of capillary refill time
SSkin color independent robust assessment of capillaryrefill time
Raquel Pantojo de Souza ∗ , George C. Cardoso Departamento de F´ısica, FFCLRP, Universidade de S˜ao Paulo, Av. Bandeirantes 3900,Ribeir˜ao Preto, SP 14040-901, Brazil
Abstract
Capillary refill time (CRT) is a method of measuring the peripheral perfu-sion status through a visual assessment. We have developed a robust systemto measure CRT using a webcam (RGB) that monitors polarized light re-flected on the skin. The average image intensity changes exponentially intime after the application of compression. The characteristic time constantwe designated polarized capillary refill time (pCRT). We tested the pCRTperformance in twenty-two healthy adult volunteers with different ages 37 ±
16, by momentarily applying a low compression (7 kPa) to the forearm ofthe volunteers. Results show positive behavior in conditions including; poorperfusion, high and low concentration of melanin, anemic conditions, anddamaged skin. In 80% of unique measurements, the result is within ± Keywords:
Capillary refill time, Peripheral perfusion, non-invasivemonitoring,
1. Introduction
Evaluation of the color of the skin is important for therapeutic and di-agnostic applications. Several studies related to the importance of colorperception in the skin; indicate the level oxygenation and peripheral per-fusion [1, 2, 3], auxiliary in diverse diagnoses, for example, venous ulcers, ∗ Corresponding author
Email addresses: [email protected] (Raquel Pantojo de Souza), [email protected] (George C. Cardoso) a r X i v : . [ phy s i c s . b i o - ph ] F e b nd neuropathy diabetic [4, 5], and others. There are many techniques forevaluating peripheral perfusion, including Capillary refill time (CRT) is amethod used to estimate peripheral perfusion status [6, 7]. CRT is definedas the time required for a distal capillary bed to regain its normal color afterhaving received enough pressure to cause blanching [8, 9] of the skin surface.This test is taken from the fingertip [10], the sternum [11], or the sites onthe forearm [12] or knee [13, 14].Measurements of CRT have reported in other studies, in intensive careunits, such as shock [15], dehydration in children [16], and the evaluation ofdengue [17]. However, these studies obtained the time, with the traditionalmeasurement CRT. In this case, the evaluation relies on the manual timeestimate and is affected by external factors with, for example, the lightingtemperature [18] of the limb and the environment, the application pressure[9] and duration are poorly standardized. Also, manual measurement of CRTrelies highly on the performance of examiners, and causes high variability inthe results [6, 19, 20]. These limitations influence the assessment of CRT forthe manual measurement were due to the lack of a gold standard.CRT test is widely acknowledged of the most tests used to evaluate pe-ripheral perfusion [21, 22]. Although a lack of robustness in the CRT mea-surements can significantly influence the measured time [6, 9, 21]. Previousstudies have relied on the CRT test involves the visual inspection, and manualtime measurement introduces intra and inter-observer variability [6, 9], fur-thermore, little is known about how this affects measure CRT in dark-skinnedpatients. Also, these studies used right compression in the skin surface andimpossibility measurement CRT in sensitive skin with low compression [9].To avoid these challenges, Blaxter et al. proposed the automated CRT de-vice that was studied using an optical assessment of diffuse reflectance [12].Other studies used to video cameras to CRT simulators for personnel train-ing [23]. However, these studies have not evaluated the dark-skinned patientsand high compression in the skin surface.Therefore, the current study focuses in measure the CRT using videoanalysis with polarized light. For that, forearm videos were used to makean image-analysis to calculate the time taken for the forearm color to returnto its baseline state after released from compression. These procedures wereperformed with polarized light and camera techniques for recording video.CRT measurement was obtained in adult healthy volunteers with differentphotoypes skin and low compression (7 kPa) to evaluate the precision andthe reproducibility of the technique. 2he purpose of our study was to develop acquisition of the video analysisfor CRT using polarized light. Was designed this decay time pCRT (PolarizedCapillary Refill Time) because it comes from the image formed by polarizedlight reflected from the skin’s surface.
2. Material and Methods
Twenty-two healthy adult people (Table 1) without a history of cardio-vascular disease participated in this study after passing a screening interview.The studies were conducted following the Declaration of Helsinki and receivedthe approval of the Medical School Clinical Hospital Ethical Committee ofthe University of S˜ao Paulo (CAAE 95342518.1.0000.5407, report number3.046.098/FFCLRP). All participants provided written informed consent be-fore the initiation of the study procedures.
Table 1: Characteristics of volunteers without know heart problems. The reported valuesare in average standard derivation (SD). Body Mass Index (BMI), Aortic pressure: SystolicBlood Pressure (SBP), Diastolic Blood Pressure (DBP) and Heart Rate (HR). n=22 subjects (9 f, 13 m) Mean ± SD Age (years) 37 16Height (cm) 172 11Weight (kg) 75 16BMI (kg/ m ) 26 5SBP(mmHg) 123 19DBP(mmHg) 77 13HR(bpm) 74 11 Volunteers remained seated in a comfortable position in a height-adjustablechair and with the right hand resting on the right thigh while the experimentsperformed on the left forearm. The acquisition performed at a controlledroom temperature 21 ◦ C, as described by Pickard et al., after acclimatiza-tion of the subject. All acquisitions realized in a dark room to ensure thatonly polarization of light was being used.3e developed a device to produce skin blanching on the volunteer’s fore-arm (Figure 1(a)) by applying pressure. The device has a cylindrical struc-ture terminated by a Teflon rounded end, that can smoothly slide insidea hollow acrylic cylinder and can be gently set to rest on the volunteer’sforearm. Teflon ending acts as a thermal insulator to prevent temperaturechanges in the skin. Cylindrical weight has a circular area of contact withthe skin of 4 cm . The compress applied in the region of interest (ROI) ofthe volunteer’s forearms at a position 9 cm from the wrist-hand line. Figure 1: a) This schematic drawing shows the support, front view (1) Cylindrical weightstandard 4 cm tip by Teflon (2) for conducting the surface test of contact with thesubject’s forearm (3) Armrest. b) Image of device,with the lighting (4) source and thevideo camera (5) with polarized P1 and P2, c) Image of the experimental, side view A video camera (Logitech HD Pro-C910, 24 fps, resolution of 1280 x720 pixels) and the light source (LED Taschibra TKL 90 - 14 W, E27) wasattached to the fixture (Figure 1(b)). Was manually focused on the videocamera on the ROI. Was installed crossed circular polarizers in the lightsource (P1) and camera (P2) to reject the specular reflection component ofthe skin [24] (Figure 1(c)). The LED lamp was on for 15 minutes before thestart of each acquisition. The volunteers remained seated for 10 minutes,monitored by an Oximeter (Model CMS50D - USB), on the left index finger,with the arm positioned at the heart level. After 10 s of video acquisition,the ROI was compressed for 5 s [9, 25] by the cylinder exerting a pressureof 7 kPa. Five repetitions were done for each volunteer, with a 1-minuterest between measurements. Videos were processed in MATLAB (MATLAB2015a, The MathWorks, Inc., Natick, Massachusetts, United States). Some4ideos excluded from the analysis because of the movement of the volunteersduring image acquisition.
Figure 2:
Image of a volunteer’s forearm with ROI region. a) During compression,b) Change in skin color after release.
The main focus of this work was to calculate the time pCRT of images ofthe forearm (Figure 2), during the color change after the pressure applied.The recorded videos were used thereafter in a separate setting to calculatepCRT via MATLAB. The averaged color of the ROI was extracted fromthe digital video file, and the color change was represented as RGB wave-forms. Figure 3 shows the curve of the average contribution of each channelRGB collect from a sequence of 35 seconds at three regions: before, duringcompression, and after the applied pressure.After the removal of the cylindrical weight standard from the skin surface,produce a pronunciation change in the intensity of the channel color. It isnotable, in the green channel, the presence of an expressive peak followed bya decay. The time required skin color to return after the pressure exertedcorrelates with the time it takes for the intensity to return to its natural state(time of decay of the exponential curve). So, was designed this decay timepCRT (Polarized Capillary Refill Time) because it comes from the imageformed by the light polarized from the skin’s surface.A curve fitting the returning phase of the green channel was modeled asan exponential decay until a critical time because the curve does not agreewith an exponential model for long times and is often noisy after about8 seconds of decay. To find the end of the region where the exponential isvalid, a sixth-order polynomial function order and a simple exponential decayare simultaneously fit to the experimental curve, the point where these twofunctions maximum divergence, we call critical time ( t c ), indicate for arrowin Figure 3b. So, we used a second exponential decay model until a critical5 igure 3: Contribution of each channel (R-Red,G-Green and B-Blue) during35 s of the CRT test before. a) The camera is locked for approximately 7 s, the extra2 s (indicated by the arrows) are for the movement of the cylindrical rod, which stopspressing for approximately 5 seconds. b) Solid line: Exponential fit between the start ofthe capillary refill critical time t c . time to determine the pCRT characteristic time, by the equation: I = I O exp ( − t/pCRT ) . (1)Data are presented as mean ± SD, minimum, and maximum. The re-producibility test was evaluated by video assessment pCRT, each video wasacquired five times. In order to evaluate fit curve, was considered uncer-tainty σ i = 0 .
05 (CI 95%) where i is each of five repetitions. Data analysiswas performed using OriginPro, version 9, and MATLAB.
3. Results
The moment the skin surface compresses, the blood is temporarily blockedout of the ROI, causing the skin to appear whitened (in some cases yellow-ish) instead of pink or brown (normal color). In healthy tissue, color returnsquickly as blood re-establishes itself in dermal capillaries [4] this color re-turn is the basis of the CRT test. The green channel exhibit the lowest6ignal-to-noise ratio and high contrast compared with other channels (Fig.3). And possibility the study the decay time of behavior in various groupsphototypes. Table 2 summarizes the value pCRT by method image analysisin each phototypes group (Figure 4). All group is possible measurement thecapillary refill time.
Table 2: Comparison value mean pCRT between different Phototype by the Fitzpatrickclassification. pCRT indicates capillary refill time by polarized measured by our inves-tigation. SD indicates average standard derivation of five repetitions measurement ofpCRT. Mean pCRT (s) ± SD Number of volunteers
Phototypes I - II 4.0 0.7 9Phototypes III - IV 4.4 1.3 8Phototypes V- VI 3.7 1.7 5In dark-skinned people (skin phototype V and VI) the change in skincolor after applied compression has low contrast, unlike what happens forlight-skinned people, skin phototype I, (Figure 4.a), making visual testingimpossible, as already reported by some authors [4, 26]. However, when weperform of the pCRT analysis (Table 2). Thus, pCRT is a method that makesit possible to quantify the CRT for different skin phototypes.
We tested the behavior of the proposed method by five repeats with oneminute between each video. For this, we tested the uncertainty calculatedby the rate σ i /pCRT i where σ considered the 95% confidence interval (CI)by adjusting the exponential green channel for each pCRT value and i isthe number of measurements for every five repetitions, this percentage rate(Figure 5).The obtained results for the proposed method in a histogram form arepresented in Fig. 5.(a). The mean of the pCRT was 4.4 ± igure 4: Image of the forearm of three volunteers with different skin pho-totypes group. a) Phototype I-II, b) Phototype III-IV and c) Phototype V-VI. Above:demarcated region shows the color of the arm before the pressure applied and below. After:releasing the compressor instrument.
The results show that in the five successive measurements CV was low(20% representing 0.5 s), only one volunteer showed a variation higher than28%. Using a single measure, with one video image, the of value pCRT wasused by the 95% CI of the exponential decay adjustment. In this investiga-tion, we have arbitrarily defined a criterion where if the uncertainty (95% CI)of the exponential decay is greater than 10% of the pCRT value; we adviseyou to retake the test. Figure 7 shows the application of this criterion. Thus,it excluded about 20 % of the measurements.In Figure 7, show the value for individual pCRT normalized by the meanpCRT, of the five repetitions for each volunteer. The rejection of the pointswith the coefficient of variation (CV) higher than 10% in the exponentialfit has produced value the confidence in one measurement of 80% of theremaining measurements are within ( ± δ = ±
20 % by pCRT normalized by the mean. If thetest performed only once, there is an 80 % chance that the result is within ±
20% of the expected pCRT value. However, the 10 % uncertainty in thefit criterion is more efficient to eliminate the overestimates of pCRT thanunderestimates. 8 igure 5: a) Distribution of N = 110 pCRT values (5 points from 22 volunteers).The trend of the points follows a Gaussian distribution (Line in red) µ = 3 . σ =1 . F W HM = 3 .
3. b) Distribution of the uncertainty evaluation by Equation: σ i /pCRT i .In both Histogram graph one point was excluded because show outside value in mean.
4. Discussion
This study considered the report of some authors [4, 26], furthermorethe lack of research that performs the traditional capillary filling time test inpatients with phototypes V and VI. As reported by [4, 26, 27], in patients withdark skin (phototypes V and VI) is difficulty in making a visual assessmentof the skin because of the high concentration of melanin in dark skin makesthe diagnosis of peripheral perfusion. The inability to visualize the colorchange in dark-skinned people make it difficult to diagnoses not only theCRT test but also in another diagnosis, as e.g. the assessment of the pressureulcer stage [4] and there is interest in areas such as assessment of tissuebreakdown leading to foot ulceration for patients with diabetes [28]. Thus, inseveral countries with high heterogeneous group colors skins, such as Brazil,its possibilities this measured the test CRT with significant variations in skincolor.This study has developed a method to analyze the image to forearmusing polarized light during the test CRT. The results showed the possibilityof measuring pCRT in volunteers with different skin phototypes and withlow compression (7 kPa). Analyzing the color change of the forearm surface9 igure 6:
Distribution of 110 pCRT values by five repetitions by each of twenty-two volunteers before the data rejection analysis.
Arrow show point the volunteer’shigh divergence. Inset show zoom of region littler variation of the coefficient of variationequal 20% Error bars show the CI 95% for the pCRT from fit curve the green channel. using polarized light with the CRT test provides the use of the techniquein different body sites. Techniques optical use transmitted light can restrictthe measurement location because it applies in regions with low deep, e.g.fingertip [29, 19]. In this work, we showed the use of circular polarizersfor image acquisition, providing the study and quantification of pCRT insubjects with skin phototypes V and VI and with low compression pressure.In these subjects, there is a high concentration of melanin in the epidermis,which causes an error in obtaining the CRT measurement, since the melaninabsorbs much of the light, so the reflected light contribution comes from themelanin contribution and not from the perfusion change caused in the dermisduring compression, this generates an error in the quantification of CRT indark skin. Polarizers eliminate a specular reflection component that allowsus to study deeper regions of the skin surface [30, 31]. For people with skin10 igure 7:
Distribution of pCRT values weighted by the average of each volun-teer before ( σ i /pCRT i ) - 110 measurements and after the data rejection analysis ( σ i /pCRT i < - 86 measurements. The inset shows that e.g. 80% of the pCRTreadings are within 20% of pCRT mean by five repetitions ( < pCRT > ) for a given vol-unteer. The gray box indicates a range of ±
20% of relative precision. The median andminimum and maximum interquartile ranges are shown. phototype V and VI, there is a high concentration of melanin on the skinsurface or difficulty in analyzing changes in skin color using CRT by thevisual method. As seen, our results showed the possibility of quantitativemeasurement pCRT in interviews with different phototypes.pCRT value gets in this work presents the same order of magnitude es-tablished in the literature for CRT. But, given the high sensitivity of RGBcameras compared to the human eye, pCRT has an acute sensitivity com-pared to traditional CRT. But these data offer a reliable reference sheds newlight on the accuracy of investigation devices for measuring peripheral per-fusion status. All subjects studied the behavior curve of the green channel11howed a better signal-to-noise ratio (SNR) compared to the other channels.With this an exponential function adjusted in this way, the decay time of theexponential considered the return time of the skin color (pCRT) for that, itwas necessary to adjust functions and thus we created a cutoff criterion ( t c )of the region of the short had more stability (Figure 3). We are not aware ofearlier studies that performed the CRT test on patients in this group (Table2). Although the study had few volunteers, the results were reproducible(Figure 7).Several factors can influence the CRT test and its results. The quantityof compression in skin surface is important for the blanching. In the tradi-tional CRT assessments, visual and manual, the compression is subjective.Kawaguchi et al. investigate the quantity pressing in the fingertips with 10kPa–70 kPa for 2 s appears to be optimal [32]. Some studies proposed thepressure using was 17 kPa [12, 33], others studies used compression in order60 kPa [19]. In this study, the results showed the possibility of measure-ment of the CRT test using compression with 7 kPa on the skin surface atthe forearm. This results in using the circularly polarized light. This tech-nique improves image resolution and extracting light that has propagatedonly within superficial tissue [24, 30, 31]. The value of pCRT showed highprecision and reproducibility.Green channel presented (Figure 3) a better signal-to-noise ratio (SNR)and greater contrast about the other channels. Two factors may be related tothis result. Disproportion is caused by the concentration of oxyhemoglobin( HbO ) in the epidermal tissue during compression. At this time, there isa decrease in HbO in the tissue and after the release of the compressioninstrument, there is an increase in the equilibrium point. Absorption spectraof oxyhemoglobin show two high absorption peaks 540 nm and 577 nm. Apossible explanation for the high variation about the baseline in the greenchannel is the fact that the sensitivity curve of the camera used in the experi-ment presents the green channel within the high sensitivity range of the HbO peak. Also, the use of a circular polarizer with a quarter-wave plate from125 nm, ensures circular polarization in the 500 nm range, thus generatingthe response in the green channel.The influence of variation skin temperature during the acquisition of thevideos and the concern of the pressure instrument to generate the presenceof pain may have caused a stressful situation in the volunteers. The stresssituation contributes to the activation of the cascade activates the sympa-thetic nervous system during the test CRT [22, 34]. Also, we believe the12ignificant increase or decrease in HR data collection may be responsible forthe high variability in inter-participant [35] in the five repetitions the testpCRT found in this study. Thus, we point one of the limitations of this ex-periment, the fact that we do not measure the temperature gradient of theskin surface during data collection. Because of the small number of volun-teers in this study, we did not investigate CRT varied with HR. However, therelationship between these factors should be analyzed in future studies.This study opens up the possibility of further work that uses the CRTtest in patients for the evaluation of peripheral perfusion and other diag-noses. Also, the development of new ways of extracting the image from theRGB channels for discoveries. Circular polarizer analyzer filtering in theacquisition of images shows the possibility of study in patients with differ-ent skin phototypes and age. Our results indicate evidence consistent withthose found by other studies [19]. However, we show a new methodologyfor image acquisition and consequently quantification of capillary refill time.Contributions show it might be possible to devise and implement a system fordetermining a robust metric to assess skin perfusion that is gentle, observer-independent, and allows for the assessment of CRT in patients with differentskin colors.
5. Conclusion
More and more the development of metrics for the assessment of periph-eral perfusion is necessary, however, the shortcoming of traditional methodsof analysis restrict its development. Recently, the use of images has shownadvantages in the processing of medical data and has achieved good effec-tiveness in many fields. Although more studies are needed to validate thesemetrics.The CRT is used to evaluate peripheral perfusion. However, the tradi-tional measurement has high uncertainty is the dependency on the observer’sperformance [6]. Therefore, in this paper, we developed a methodology us-ing light polarized for image acquisition and processing to be used in themeasurement of peripheral perfusion.The green channel decay showed a good contrast that allows the deter-mination of a characteristic time (pCRT) that is a surrogate for CRT. Thismethodology was able to quantify the CRT in volunteers with various groupsprototypes skin, with a high concentration of melanin, in anemic conditionswith low compressure (7 kPa). Experimental results show that the pCRT13odel has achieved high accuracy (approximate 80%), and has the potentialand value to promote the clinical application.
6. Acknowledgements
The authors special thanks to Beatriz Janke, C. Renato da Silva and allthe volunteers who participated in the experiment. This study was financedin part by the Coordena¸c˜ao de Aperfei¸coamento de Pessoal de N´ıvel Superior- Brasil (CAPES) - Finance Code 001.