Eduard Kieser

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/- Development of diagnostic sensors for infant dehydration assessment using optical methods Conference Paper · August 2015 DOI: 10.1109/EMBC- CITATION READS 1 111 5 authors, including: Kiran H Dellimore Eduard Kieser Philips Stellenbosch University 64 PUBLICATIONS 180 CITATIONS 5 PUBLICATIONS 1 CITATION SEE PROFILE SEE PROFILE Cobus Visser Johan Smith Stellenbosch University Stellenbosch University 4 PUBLICATIONS 1 CITATION 67 PUBLICATIONS 532 CITATIONS SEE PROFILE SEE PROFILE All content following this page was uploaded by Kiran H Dellimore on 10 September 2015. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document and are linked to publications on ResearchGate, letting you access and read them immediately. Development of diagnostic sensors for infant dehydration assessment using optical methods Eduard Kieser, Kiran Dellimore, Member IEEE, Cornie Scheffer, Member, IEEE, Cobus Visser, and Johan Smith  Abstract— Dehydration resulting from acute diarrhea is one of the leading causes of infant mortality in the developing world. Safe assessment of an infant’s hydration level is essential to determine appropriate clinical intervention strategies. However, clinical hydration scales, which are the current gold standard for non-invasive hydration assessment, are often unreliable in lower resource settings. This study presents the development and testing of non-invasive, optical sensors for the objective assessment of dehydration based on the quantitative measurement of skin recoil time, capillary refill time and skin temperature. The results obtained have demonstrated the basic feasibility of using optical sensors for the objective assessment of dehydration. However, several challenges must be overcome before these sensors can be applied in a clinical setting. D I. INTRODUCTION ehydration resulting from acute diarrhea is one of the leading causes of infant mortality, resulting in over 1.3 million infant deaths each year worldwide [1]. The current gold standard for prospective dehydration assessment involves the use of hydration scales, which aggregate various subjective clinical observations into a score [2-6]. Among the most commonly used markers of dehydration include: abnormal breathing pattern, capillary refill, eye and fontanelle appearance, heart rate, level of physical activity, mouth dryness, neurologic state, pulse quality, skin turgor, absence of tear production, thirst, temperature difference between core and extremities, as well as urine color and smell [2-7]. The three most commonly utilized hydration scales which are based on a subset of these dehydration markers include the Clinical Dehydration Scale (CDS), the WHO dehydration scale, and the 4-point and 10-point Gorelick hydration scales [3,5]. However, previous work has shown that these scales are ineffective in assessing dehydration in lower resource settings [3,6], where the vast majority of infant deaths occur due to dehydration caused by diarrheal diseases [1]. This may be attributed in part to the lack of training and inexperience of clinical personnel in these settings, as well as the subjectivity and lack of standardization of the tests for various markers [3]. Hence, there is a proEduard Kieser, Kiran Dellimore, Cornie Scheffer, and Cobus Visser are with the Biomedical Engineering Research Group, Department of Mechanical and Mechatronic Engineering, Stellenbosch University, South Africa. (phone: -; fax: -; e-mail:-. Johan Smith is with the Department of Paediatrics & Child Health, Tygerberg Hospital & Stellenbosch University, South Africa. -/15/$31.00 ©2015 IEEE found need for more objective and quantitative, as well as user-friendly methods of dehydration assessment which are effective in resource constrained settings. Skin turgor and capillary refill time are two of the most reliable non-invasive individual markers of infant hydration [89]. The presence of cool extremities has also been found to be a useful indicator of dehydration [10]. Many studies that have investigated non-invasive hydration markers have concluded that scores based on a combination of hydration markers consistently outperformed any single indicator [2,8,11]. Most markers for dehydration are continuous parameters, however, because they are difficult to quantitatively assess clinically, they are reduced to categorical values when used for hydration assessment. Previous attempts at objectifying dehydration assessment using non-invasive optical sensors include work by Shavit et al. [12], Brodoly et al. [13] and Kviesis-Kipge et al. [14]. These studies all investigated the objective measurement of capillary refill time (CRT) using various optical techniques including digital videography and light spectrometry. To date there is a dearth of literature on the non-invasive measurement of skin recoil time (SRT) and skin temperature (ST), in the context of dehydration assessment. The aim of this study is to present the development and testing of novel diagnostic tools for the objective assessment of dehydration in infants in lower resource settings based on three different non-invasive optical sensors, which quantitatively measure SRT, chest CRT and ST. II. METHODS A. Design and implementation of diagnostic sensors Among the most important design requirements considered during development of the three dehydration assessment sensors are that they should be: low-cost, non-invasive, portable, quantitative, reliable, and user-friendly. The development or selection of each sensor also necessitated additional sensor specific considerations. Capillary refill time sensor The CRT sensor was designed to objectify the capillary refill test which is manually performed by clinicians. To accomplish this it is necessary to control the magnitude and duration of the applied pressure. A diagram of the main components of the CRT sensor is shown in Fig. 1. The sensor has a silicone pressure pad that is mounted on a swing arm, over a Honeywell FS1500 force sensor which is connected to the controller via a TI 1115ads 16bit analogue to digital converter. When the swing arm is locked into place it 5537 Skin temperature profile sensor The ST sensor was selected to test the hypothesis that an irregular skin temperature profile is linked to dehydration. A FLIR-E60 thermal camera (FLIR Systems, Wlisonville, OR, USA) was used to measure skin surface temperature over a prescribed path on the subject‟s body (Fig. 3). The thermal camera has a sensitivity of 0.05°C, an accuracy of less than 2 °C and acquires images with a resolution of 320x240 pixels. Fig. 1: CRT sensor concept and the key components used to build the device: 1&2 - Conceptual layout of the device, 3 - Photograph of the CRT sensor, 4 - Camera module, 5 - Analogue to digital converter, 6 Force sensor, 7 - Lighting and sensor interface board, 8 - Color enhanced view of blanching image acquired by the CRT sensor. can be used to apply a blanching pressure while the measured force is output on a 1.2 inch tri-color 8x8 LED Matrix display to ensure that the applied force falls within the prescribed limits (3-3.5N for infants and around 10N for adults). When the force has been applied for 6 seconds the operator can pull a lever that allows the swing arm to swing out of the way which allows the 5MP Raspberry-Pi camera (set to 1024x768 pixels @ 30fps) to start recording the scene. The scene is illuminated by four white LEDs controlled by the lighting board shown in Fig. 1. The resulting recordings are then saved to a flash-drive for off-line analysis. The sensor is powered by two 3.7 V 2000 mAh lithium polymer cells which are connected in series and controlled by a 5 V linear regulator. To enable the camera to focus at the required length, a 10 mm diameter plano-convex lens with a focal length of 30 mm, was installed in front of the camera. Fig. 2: 1 - Photograph of the SRT sensor, 2 - Application scenario, 3 Raw SRT sensor images. Skin recoil time sensor The main design challenge with the skin recoil time sensor is ensuring that it has good time resolution. To facilitate this, the skin recoil time sensor was developed using the same raspberry-pi camera (OmniVision OV5647) as used in the CRT sensor, however, in the SRT sensor it is set to record at a resolution of 640x480 pixels and 90 fps. The camera sensor is mounted on a movable head with two degrees of freedom to facilitate operation. The camera‟s focal length was also shortened by using a 30 mm plano-convex lens with a very narrow field of view and focus range. A highly directional lighting system was then implemented using two LEDs to ensure that the user could easily see whether the sensor is at the correct orientation and distance. The sensor was designed to rest on the patient‟s sternum while the physician performers a normal skin recoil test on the abdomen of the patient as depicted in Fig. 2. Fig. 3: (a) Sample ST image (b) FLIR E-60 thermal camera. B. Testing Procedure The performance of the three sensors was evaluated by conducting two separate in vivo clinical studies, one on adults and one on infants. Prior to initiating the clinical studies the in-house built devices were validated by performing preliminary CRT and SRT tests in a laboratory setting. The CRT and SRT sensor data were compared to manual data obtained by two physicians on the 3 healthy adult male subjects. The good correlation between the measurements was used to confirm that the devices performed as expected. Both clinical studies were approved by the Stellenbosch University Health Research Ethics Committee (#S13/10/204, “Quantitative Hydration Sensor Development Adult/Infant Testing”). Informed consent was obtained from the adult volunteers and from the parents of the infant participants before being enrolled in the study. A total of 9 subjects were recruited for the adult study including 5 males and 4 females (age = 23.4±1.59 years, BMI = 24.9±3.3). The adult tests were performed over a period of 6 days, with all exercise sessions conducted at the Stellenbosch University Sport Performance Institute under the guidance of a sports trainer. The first three days were used to determine the baseline euhydrated parameters for the patients. In the last three days the volunteers exercised for four sessions of 50 minutes each day. Each training session was followed by a 10-20 minute cool down period after which the hydration markers were measured. The infant study was conducted at the Paediatric Ambulatory Admission Unit at the Tygerberg Children‟s Hospital. Subjects were eligible for enrolment if they were admitted with diarrhea, presented with signs of mild to severe dehydration as determined by an attending physician, were between the ages of 6 and 36 months and did not present with signs of severe malnutrition (i.e., their weight was not more than 3 standard deviations away from the mean weight-forage based on WHO weight-for-age-charts [15]). A total of 10 infants were enrolled in the study (6 male and 4 female, with a mean age of 11 months). Once enrolled all patient information was recorded as well as one “snapshot” of the patient condition. A snapshot consists of: (i) the clinical as- 5538 sessment recorded by the physician which included an assessment of CRT, SRT, eye and fontanelle appearance, mucous membranes dryness, neurological state, tear production and pulse quality and (ii) the nurse/digital assessment which included standard patient data such as blood pressure, pulse rate, body temperature and body weight as well as an assessment of ST, CRT and SRT using the three sensors. C. Data Analysis The first step in extracting the CRT from the raw images involved the marking of the center of each blanching site at the start of each recording. Since the blanching sites often appear very faint, an exaggerated view of the image using subsequent image information to facilitate the blanching exaggeration was used (Fig. 1). Once all the start positions were marked, the color of the blanching site was logged as it changed with respect to time. Due to infant restlessness optical flow tracking methods were applied to ensure that the correct part of the scene was used as the blanching site. A Shi-Tomasi corner detector was used to find corners in the image and the Lucas-Kanade method was employed to match the points of interest to the corresponding points in subsequent frames [16]. This information was then used to move the region of interest markers with the patient‟s skin. Due to excessive motion and non-uniform lighting conditions most of the data were rejected. The extracted color information was then plotted with respect to time to allow manual CRT selection to be performed to ensure that the selected times were not compromised by unexpected artifacts in the color data. SRT was determined from the raw data by manually marking the start and stop time of each recoil test by scrolling through the acquired images. Manual marking was used because of motion blur in the frames which made it difficult to reliably track a point on the skin automatically, using computer vision methods. Fig. 2-3 shows an example of raw data from of a dummy test conducted on the fore arm. To extract a hydration marker from the raw thermal images a function was used to mark a path on the thermal images as shown by the solid line in Fig. 3(b). The program would then extract the patient‟s temperature along the specified paths. The parameters extracted to assess dehydration were the difference between the highest and lowest temperature on the path, the number of local maxima/minima on the path as well as the largest temperature gradient along the path. In all cases manual inspection of the extracted data was used to determine the data integrity. A linear regression was performed where the observed dehydration (as determined by post illness weight gain) was treated as the outcome variable and CRT, SRT and maximum temperature gradient, as well as pulse and breathing rate were used as the input variables. This was used to predict the expected hydration state as a function of the chosen parameters. Unlike in the adult study the clinical and digital assessments for the infant study were not done synchronously. Therefore the forms were consolidated into one inclusive data structure using the time and date stamps as matching metrics. The difference in time between matched entries was stored and all matches with measurement intervals of more than 2 hours were rejected. III. RESULTS Fig. 4 shows the grouped data for the two clinical dehydration studies. The „x‟ and „●‟ symbols correspond to the adult and infant data, respectively, while the solid line indicates the regression line fit to the data. In Fig. 4 (a) CRT is plotted against observed dehydration. The figure shows an inverse relation between CRT and observed dehydration level for adults (p < 0.001 and R2= 0.147), while for infants no clear correlation is observed (p = 0.388 and R2 = 0.023). Average CRT values of 3.10 ±1.55 s and 2.00 ± 1.04 s, were obtained for the adult and infant studies, respectively. In Fig. 4 (b) SRT is plotted against observed dehydration. The adult data shows an inverse relationship between SRT and weight change (p < 0.001 and R2= 0.129), while for the infant data SRT shows a positive correlation with dehydration level (p < 0.001 and R2= 0.247). The SRT for the adult population was on average longer compared to the infant population with mean recoil time of 0.12 s and 0.06 s, respectively. As a binary predictor for 5% dehydration in infants the SRT had a sensitivity of 0.80 and a specificity of 0.84 when 0.08 s was used as a predictor threshold value (Fig. 5 (b)). Figure 4 (c) shows the maximum temperature gradient on a path across the abdomen for the infants and on a path across the cheek for the adult study plotted with respect to hydration level. For the adult study a negative relationship is observed (p = 0.015 and R2 = 0.055) and the infant study shows a weak positive correlation (p = 0.040 and R2 = 0.096) between lev- Fig. 4: (a) CRT, (b) SRT and (c) Maximum temperature gradient as a function of observed dehydration (%) for the: (left) adult study and (right) infant study. el of dehydration and maximum temperature gradient. Fig. 5 (a) shows the predicted dehydration, as calculated by the linear regression model, plotted against the observed 5539 dehydration level, for each snapshot in the infant study. It is not the straight line one would see in a perfect sensor, however one can see a strong correlation between the predictor and the reference with R2 = 0.573 and p < 0.001. The receiver operating characteristic (ROC) curves for determining a dehydration level of 5% in infants is shown in Fig. 5 (b). The „●‟, „■‟, „▲‟, and „♦‟ symbols correspond to the SRT, CRT, maximum temperature gradient and the fusion model respectively. The areas under the ROC curves (AUC) for the SRT sensor, the CRT sensor, the thermal profile sensor, and the combined model are 0.8, 0.52, 0.73 and 0.86 respectively. The best sensitivity for the combined approach is 1 with a specificity of 0.79. was the temperature dependence of the dehydration markers which was linked to the study design. In the infant tests motion artifacts due to patient restlessness and overburdened hospital staff (which led to large time differences between entries) limited the data quality obtained. V. CONCLUSION The results obtained have demonstrated the basic feasibility of the three non-invasive optical sensors based on CRT, SRT and ST, for the objective assessment of dehydration status. Despite the small sample sizes of the adult and infant in vivo studies, the results showed that by combining the three sensor measurements into a composite index it is possible to achieve an objective measure of dehydration level with AUC of 0.86, sensitivity of 1.0 and specificity of 0.79. Future work will focus on improving sensor performance and reliability over a wide range of clinically relevant conditions. VI. ACKNOWLEDGMENT Gratitude is expressed to THRIP and Philips Research for financial support, and to the staff at Tygerberg Hospital. REFERENCES Fig. 5: (a) Predicted dehydration (regression model) vs. observed dehydration for the infant study. (b) ROC for the individual sensors and the sensor fusion approach for the infant study. IV. DISCUSSION Fig. 4 (a) shows very little correlation between the digital CRT and the observed dehydration in the infants. This is due to artifacts resulting from excessive motion of the infants during the tests, which resulted in 60% of the infant CRT data being rejected. However, previous CRT studies have found stronger correlations between dehydration and capillary refill time (sensitivity and specificity of 0.60 and 0.85 compared to 0.50 and 0.50 in the present study) [8]. In the adult study both the SRT and CRT sensors showed correlations that were inconsistent with expectation since these quantities showed an inverse relationship with observed dehydration level (Fig. 4 (a) and (b)). This is likely due to a temperature dependence linked to the study design [11]. The participants were always well hydrated in the mornings when it was cooler and less hydrated in the afternoons when it was warmer. Use of the maximum temperature gradient as a dehydration indicator, Fig. 4 (c), yielded reasonably promising results (AUC: 0.73, sensitivity: 0.68, specificity: 0.59). The digital sensor fusion approach shown in Fig. 5 (a) and (b) shows that by combining the 3 different markers using continuous values for the predictors and linear regression achieved better dehydration prediction performance than any of the individual indicators (AUC: 0.86, sensitivity: 1.0, specificity: 0.79). This corroborates the findings of previous studies that showed that scores based on a combination of hydration markers consistently outperformed any single indicator [2,8,11]. It is important to note that there were many significant challenges in the testing of the sensors which may have limited the quality of the data acquired in the current study. In the adult tests, the main confounding factor [1] M. Santosham, A. Chandran, S. Fitzwater, et al., “Progress and barriers for the control of diarrhoeal disease,” Lancet, vol. 376, pp. 63–7, 2010. [2] J. Friedman, R. Goldman, R. Srivastava, et al., “Development of a clinical dehydration scale for use in children between 1 and 36 months of age.,” J Pediatr., vol. 145, pp. 201–7, 2004. [3] K. Pringle, S. Shah, I. Umulisa, et al., “Comparing the accuracy of the three popular clinical dehydration scales in children with diarrhea,” Int J Emerg Med., vol. 4, no. 1, pp. 58, 2011. [4] L. Kinlin and S. Freedman, “Evaluation of a Clinical Dehydration Scale in Children Requiring Intravenous Rehydration,” Pediatrics, vol. 129, pp. e1211–e1219, 2012. [5] J. Jauregui, D. Nelson, E. Choo, et al., “External validation and comparison of three pediatric clinical dehydration scales,” PLoS One, vol. 9, pp. 1– 6, 2014. [6] A. Falszewska, P. Dziechciarz, and H. Szajewska, “The diagnostic accuracy of clinical dehydration scales in identifying dehydration in children with acute gastroenteritis: A systematic review,” Clin Pediatr., vol. 53, pp. 1181–8, 2014. [7] L. E. Armstrong, “Assessing hydration status: the elusive gold standard,” J Am Coll Nutr., vol. 26, pp. 575S–584S, 2007. [8] T. Popov, “Review: capillary refill time, abnormal skin turgor, and abnormal respiratory pattern are useful signs for detecting dehydration in children,” Evid Based Nurs., vol. 8, pp. 57, 2005. [9] A. Canavan and B. Arant, “Diagnosis and management of dehydration in children,” Am Fam Physician, vol. 80, pp. 692-6, 2009. [10] M. J. Steiner and D. A. Dewalt, “Is this child dehydrated ?” JAMA, vol. 291, pp. 2746–54, 2004. [11] M. Gorelick, K. Shaw, and K. Murphy, “Validity and reliability of clinical signs in the diagnosis of dehydration in children,” Pediatrics, vol. 99, pp. E6, 1997. [12] I. Shavit, R. Brant, C. Nijssen-Jordan, et al., “A novel imaging technique to measure capillary-refill time: improving diagnostic accuracy for dehydration in young children with gastroenteritis,” Pediatrics, vol. 118, pp. 2402–8, 2006. [13] A. Bordoley, R. Irwin, V. Kalyani, et al., “Digit: Automated, temperature-calibrated measurement of capillary refill time,” MEAM- Senior Design Project - Final Report, pp. 1-13, 2012. [14] E. Kviesis-Kipge, E. Curkste, et al., “Optical studies of the capillary refill kinetics in fingertips,” IFMBE Conf Proc., vol. 25, pp. 377–9, 2009. [15] World Health Organization, “New child growth standards,” pp. 5, 2006. [16] B. Lucas and T. Kanade, “An iterative image registration technique with an application to stereo vision,” Imaging, vol. 130, pp. 674–9, 1981. 5540 View publication stats