|Year : 2016 | Volume
| Issue : 3 | Page : 131-135
|Digital camera image analysis of faeces in detection of cholestatic jaundice in infants
Parinya Parinyanut1, Tai Bandisak2, Piyawan Chiengkriwate1, Sawit Tanthanuch2, Surasak Sangkhathat1
1 Department of Surgery, Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla 90110, Thailand
2 Department of Electrical Engineering, Faculty of Engineering, Prince of Songkla University, Hat Yai, Songkhla 90110, Thailand
Click here for correspondence address and email
|Date of Web Publication||5-Aug-2016|
| Abstract|| |
Background: Stool colour assessment is a screening method for biliary tract obstruction in infants. This study is aimed to be a proof of concept work of digital photograph image analysis of stool colour compared to colour grading by a colour card, and the stool bilirubin level test. Materials and Methods: The total bilirubin (TB) level contents in stool samples from 17 infants aged less than 1 year, seven with confirmed cholestatic jaundice and ten healthy subjects was measured, and outcome correlated with the physical colour of the stool. Results: The seven infants with cholestasis included 6 cases of biliary atresia and 1 case of pancreatic mass. All pre-operative stool samples in these cases were indicated as grade 1 on the stool card (stool colour in healthy infants ranges from 4 to 6). The average stool TB in the pale stool group was 43.07 μg/g compared to 101.78 μg/g in the non-pale stool group. Of the 3 colour channels assessed in the digital photographs, the blue and green light were best able to discriminate accurately between the pre-operative stool samples from infants with cholestasis and the samples from the healthy controls. With red, green, and blue (RGB) image analysis using wave level as the ANN input, the system predicts the stool TB with a relationship coefficient of 0.96, compared to 0.61 when stool colour card grading was used. Conclusion: Input from digital camera images of stool had a higher predictive capability compared to the standard stool colour card, indicating using digital photographs may be a useful tool for detection of cholestasis in infants.
Keywords: Cholestatic jaundice, digital camera, image analysis
|How to cite this article:|
Parinyanut P, Bandisak T, Chiengkriwate P, Tanthanuch S, Sangkhathat S. Digital camera image analysis of faeces in detection of cholestatic jaundice in infants. Afr J Paediatr Surg 2016;13:131-5
|How to cite this URL:|
Parinyanut P, Bandisak T, Chiengkriwate P, Tanthanuch S, Sangkhathat S. Digital camera image analysis of faeces in detection of cholestatic jaundice in infants. Afr J Paediatr Surg [serial online] 2016 [cited 2019 May 26];13:131-5. Available from: http://www.afrjpaedsurg.org/text.asp?2016/13/3/131/187810
| Introduction|| |
Biliary atresia is a rare but serious disease in infants characterised by progressive sclerosing obstruction of the biliary tree that can lead to biliary cirrhosis and end-stage liver disease. ,, In high-income countries, this disease is the most frequent indication for liver transplantation in children, , but in resource-limited countries, the only surgical option is hepatic portoenterostomy (Kasai's operation).  Previous studies have reported that the chance of achieving successful biliary drainage after Kasai's operation decreased the older the patient was at the time of the operation. ,, Other studies have suggested the cut-off age to have a higher chance to have a successful Kasai outcome is 45-60 days, , thus early diagnosis is a crucial factor in preserving the lives of these infants. 
Several methods have been proposed as screening tools for early detection of infantile cholestasis, from clinical evaluation to ultrasonography and serum markers.  As pale stool is the main clinical clue of the disease, one of the earlier suggestions was using stool colour as a screening strategy. ,,, The main problem with using stool colour in this way, however, was standardisation, i.e., what level of paleness can accurately be associated with biliary obstruction. In 1994, a semi-objective quantitation system using a stool colour card was proposed by Japanese researchers.  A recent report from the same researchers of a 19-year cohort found that long-term use of the stool colour card in Tochigi Prefecture increased the proportion of patients who received Kasai's operation before 60 days of age and also improved survival with native liver when compared to the community that did not use this tool.  Interestingly, their data demonstrated a high specificity of screening by stool colour (100% at 1 month of age), which led to our research question - Is it possible to apply an image processing device in discriminating pale stool from infants with cholestasis from normal stool from well babies? The main objective of this study was to determine the level of correlation between stool colour grading by a stool colour card, colour parameters derived from a digital image, and actual stool bilirubin level in infants.
| Materials and Methods|| |
The study was approved by the Institutional Research Board of the Faculty of Medicine, Prince of Songkla University. Informed consent was obtained from a parent or guardian of every infant before sample collection. Stool samples were collected from 7 infants under 3 months old with a diagnosis of extrahepatic biliary obstruction. Samplings were performed within 1 week before the biliary drainage operation and at 1 week, 1 month, 3 months and 6 months after the operation. Each sample was divided into two pieces, first piece was graded for stool colour, photographed and another was frozen at −80°C for further stool bilirubin test. Stool samples from 10 infants under 6-month-old who attended the Well Child Clinic in our institute were also collected as a control group, and went through the same process of colour evaluation and bilirubin measuring [Figure 1].
The stool colour card used in this study was a gift from Professor Mei-Hwei Chang, Department of Pediatrics, National Taiwan University Hospital, Taiwan.  Two researchers graded the colour of stool samples under sunlight into 6 levels. Digital image acquisition was carried out in a black box, which projected one colour at a time-red light (wavelength 625.0 nm), green light (wavelength 529.0 nm) and blue light (wavelength 469.0 nm). Each stool sample was photographed 3 times, once each under each wave length, for later image analysis using the histogram mode in the Photoshop PS3 program (Adobe Systems Incorporated, San Jose, CA, USA). The wave level that gave the highest amplitude on the histogram in each colour was chosen for analysis. Before measuring stool bilirubin, stool samples were thawed, ultrasonicated, filtered and centrifuged at 14,000 rpm × 15 min at 25°C to remove insoluble particles. Quantitation of stool total bilirubin (STB) was done using the Ehrlich technique (Diazo reaction) (Stanbio Laboratory, Texas, USA).
Sample sizes were calculated for correlation coefficients of 0.5, Type I error at 0.05 and Type II error at 0.20. The calculations required 30 samples. Categorical data were analysed for frequency (per cent), and continuous data for mean and range (unless stated otherwise). Data calculations were performed using the statistical programme Stata 14 (StataCorp, Texas, USA). The artificial neural network (ANN) was created using the MATLAB 2012a program and it's Neural Network Toolbox, Feed-Forward network with back propagation leaning algorithm, log sigmoid function for hidden layer and linear function for output (MathWorks, Inc.). The ANN training used wave levels and stool colour card grade as inputs and STB as the outcome.
| Results|| |
From August 2013 to August 2014, 31 stool samples were collected from 17 subjects, seven with a confirmed diagnosis of biliary tract obstruction and 10 healthy infants who came for vaccination and check-up at our Well Child Clinic. The male:female ratio was 9:8. The mean age of the infants was 88 days (49-157 days). Of all samples, 26 (84%) were collected when the infants were exclusively fed by milk. Of 7 infants with extrahepatic biliary obstruction, six had biliary atresia and one had idiopathic pancreatic head fibrosis. All seven cases underwent biliary drainage surgery.
Stool colour grading using the colour card gave concordance between 2 observers at r2 = 0.99 by linear regression analysis and Cohen's kappa agreement test gave a value of kappa at 1.0. The stools were graded as grade 1 in 10 samples, grade 2 in 9 samples, grade 4 in 2 samples, grade 5 in 3 samples and grade 6 in 7 samples. All pre-operative stool samples from cholestatic patients gave colour grade 1 while post-operative stool samples from the same group ranged in colour grade. All stool samples from healthy infants gave a colour grade of four or more [Figure 2].
|Figure 2: Distribution of stool colour grading in 31 samples from 17 infants|
Click here to view
The average STB was 6.58 μg/mg of stool (range 0.36-25.98 μg/mg of stool). The average STB in stool samples that were in colour grades 1-2 was 4.31 μg/mg of stool, compared to 10.18 μg/mg of stool in those samples in colour grades 4-6 (Mann-Whitney P < 0.01). The distribution of STB values in cholestasis infants and healthy babies are shown in [Figure 3]. There was no statistical difference in average STB values between sexes (male 6.35 μg/mg of stool and female 6.73 μg/mg of stool, P = 0.88).
|Figure 3: Distribution plots and mean levels (+) of stool total bilirubin in cholestasis patients and healthy infants|
Click here to view
The average wave levels in the digital photos taken under red, green and blue (RGB) light were 181.94 ± 48.38, 121.83 ± 46.52 and 126.00 ± 72.70, respectively. Wave levels in all 3 colours gave significantly different values between pale stool (grades 1-3) and non-pale stool (grades 4-6) [Table 1].
|Table 1: Comparing average wave levels in digital images derived from 3 light channels between pale stool and non-pale stool*|
Click here to view
On linear regression between wave levels and STB, wave levels from RGB channels gave correlation coefficients of 0.30, 0.28 and 0.26, respectively. On comparing wave levels between pre-operative stool samples of cholestatic cases and the control samples from healthy infants, it was found that blue channel could best discriminate the pale stool from the non-pale [Figure 4]. When receiver operating characteristic curves were plotted for the wave levels of each colour using stool grading groups by colour card (pale vs. non-pale), the areas under the curve of RGB were 0.77, 0.83 and 0.80, respectively [Figure 5].
|Figure 4: Comparison of wave levels between stool samples from cholestasis infants and healthy controls under three wavelengths|
Click here to view
|Figure 5: Receiver operating characteristic curves plots between sensitivity and 1 - specificity of wave levels in the red, green and blue channels and stool colours graded by colour card (pale/non - pale). Blue - adjusted refers to the analysis with three outliers removed from the calculation on the blue - channel analysis. AUC: Area under the curve|
Click here to view
When wave level from three light colours were used as inputs in the training sets of the ANN, the three-colour combination (red green and blue) gave the best correlation between predicted STB and actual STB [Figure 6]. Using stool colour card grades as an input set, the ANN could predict STB with an R-value of 0.61 (graph not shown).
|Figure 6: Plots between stool total bilirubin values predicted by the artificial neural network and actual stool total bilirubin assessed by laboratory tests in different input models; RGB: Red, green, blue; RG: Red and green; RB: Red and blue; GB: Green and blue|
Click here to view
| Discussion|| |
Stool colour examination is a crucial part of screening for infantile cholestatic diseases, especially biliary atresia. The stool colour card is a cost-effective strategy that has been widely accepted in areas with relatively high incidence of the disease. Consistent with previous reports in Asians, this study also found that pale stool, matching grade 1/6 on the colour card, was highly specific to the infants with cholestatic diseases. In addition, none of the stool samples from normal infants had a colour grade <4/6. This study also found that although the colour grading had a very high agreement between observers, there remained some probability of disparities among graders. Objective colour assessment by digital image processing may help reduce such disparities.
Red-green-blue (RGB) digital image analysis has been applied in various areas of biomedical diagnosis. ,, Examples of reported applications are an assessment of cutaneous microcirculation,  cutaneous pigments  and bacterial identification.  Bilirubin is a yellow pigment with a light absorption in the 450-470 nm spectrum, which is closest to the blue light wavelength.  When conjugated bilirubin is excreted into the bowel, bilirubin is converted by gut into stercobilin, which has maximum light absorption at 492-496 nm. These characteristics explain our finding that blue and green light sources had the highest power in discriminating pale stool in cholestatic infants from normal stool in age-matched controls. ROC analysis also confirmed that the blue and green wave level gave the highest correlation with the stool card grouping. Interestingly, wave level parameters from all three channels could be trained to predict stool bilirubin under an ANN model. We chose to use ANN to predict stool bilirubin using colour wave level input as the correlation was in non-linear model. Predictions using RGB showed a higher accuracy than the stool colour card.
There were some limitations with using digital image capture we noted from this study. First was that stool may not be homogenous, and a spot image acquisition might not represent the colour of the whole specimen, leading to incorrect conclusions. Although the majority of our patients were infants who received only milk products, potential contamination by pigments found in some prepared foods also needs to be considered when using a highly sensitive image-based colour analysis such as in this study, and this test might not be reliable in patients old enough to be receiving solid foods. Moreover finally, the digital image photography in this study was performed in a highly controlled environment, and to be an effective screening test usable in a wide range of settings, the method should be further improved to be reliable less controlled light situations.
| Conclusion|| |
This study offers a preliminary evaluation of the feasibility of using RGB digital image analysis in diagnosing cholestatic-associated pale stool in infants. The study demonstrated a significant correlation between wave levels in all 3-colour channels and pale stool, with the blue light wave level giving the highest value in reliably identifying biliary atresia from pale stool in this group of patients.
The authors would like to thank Professor Mei-Hwei Chang, College of Medicine, National Taiwan University and the Health Promotion Administration, Ministry of Health and Welfare, Taiwan for permission to use the Taiwan stool colour card. Assistant Professor Chittanon Buranachai, Department of Physic, Faculty of Science, Prince of Songkla University helped measurement the wave lengths on the LED lamp. The Photoshop PS3 (Adobe, Inc.) used in this study was the license of computer centre of Prince of Songkla University and the MATHLAB used in this study was licensed to the Faculty of Engineering, Prince of Songkla University (License number 161052).
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Yoon PW, Bresee JS, Olney RS, James LM, Khoury MJ. Epidemiology of biliary atresia: A population-based study. Pediatrics 1997;99:376-82.
Sangkhathat S, Patrapinyokul S, Tadtayathikom K, Osatakul S. Peri-operative factors predicting the outcome of hepatic porto-enterostomy in infants with biliary atresia. J Med Assoc Thai 2003;86:224-31.
Schreiber RA, Barker CC, Roberts EA, Martin SR, Alvarez F, Smith L, et al.
Biliary atresia: The Canadian experience. J Pediatr 2007;151:659-65, 665.e1.
Wildhaber BE. Biliary atresia: 50 years after the first Kasai. ISRN Surg 2012;2012:132089.
Chardot C, Carton M, Spire-Bendelac N, Le Pommelet C, Golmard JL, Auvert B. Prognosis of biliary atresia in the era of liver transplantation: French national study from 1986 to 1996. Hepatology 1999;30:606-11.
Nio M, Ohi R, Miyano T, Saeki M, Shiraki K, Tanaka K; Japanese Biliary Atresia Registry. Five- and 10-year survival rates after surgery for biliary atresia: A report from the Japanese Biliary Atresia Registry. J Pediatr Surg 2003;38:997-1000.
Sokol RJ, Shepherd RW, Superina R, Bezerra JA, Robuck P, Hoofnagle JH. Screening and outcomes in biliary atresia: Summary of a National Institutes of Health workshop. Hepatology 2007;46:566-81.
Serinet MO, Wildhaber BE, Broué P, Lachaux A, Sarles J, Jacquemin E, et al.
Impact of age at Kasai operation on its results in late childhood and adolescence: A rational basis for biliary atresia screening. Pediatrics 2009;123:1280-6.
Balistreri WF. Early diagnosis of biliary atresia. J Pediatr 2015;166:783-7.
Wang H, Malone JP, Gilmore PE, Davis AE, Magee JC, Townsend RR, et al.
Serum markers may distinguish biliary atresia from other forms of neonatal cholestasis. J Pediatr Gastroenterol Nutr 2010;50:411-6.
Chen SM, Chang MH, Du JC, Lin CC, Chen AC, Lee HC, et al.
Screening for biliary atresia by infant stool color card in Taiwan. Pediatrics 2006;117:1147-54.
Hsiao CH, Chang MH, Chen HL, Lee HC, Wu TC, Lin CC, et al.
Universal screening for biliary atresia using an infant stool color card in Taiwan. Hepatology 2008;47:1233-40.
Lien TH, Chang MH, Wu JF, Chen HL, Lee HC, Chen AC, et al.
Effects of the infant stool color card screening program on 5-year outcome of biliary atresia in Taiwan. Hepatology 2011;53:202-8.
Gu YH, Yokoyama K, Mizuta K, Tsuchioka T, Kudo T, Sasaki H, et al.
Stool color card screening for early detection of biliary atresia and long-term native liver survival: A 19-year cohort study in Japan. J Pediatr 2015;166:897-902.e1.
Matsui A, Ishikawa T. Identification of infants with biliary atresia in Japan. Lancet 1994;343:925.
Nishidate I, Tanaka N, Kawase T, Maeda T, Yuasa T, Aizu Y, et al.
Noninvasive imaging of human skin hemodynamics using a digital red-green-blue camera. J Biomed Opt 2011;16:086012.
Nishidate I, Wiswadarma A, Hase Y, Tanaka N, Maeda T, Niizeki K, et al.
Noninvasive spectral imaging of skin chromophores based on multiple regression analysis aided by Monte Carlo simulation. Opt Lett 2011;36:3239-41.
Nishidate I, Maeda T, Niizeki K, Aizu Y. Estimation of melanin and hemoglobin using spectral reflectance images reconstructed from a digital RGB image by the Wiener estimation method. Sensors (Basel) 2013;13:7902-15.
de Almeida VE, da Costa GB, de Sousa Fernandes DD, Gonçalves Dias Diniz PH, Brandão D, de Medeiros AC, et al.
Using color histograms and SPA-LDA to classify bacteria. Anal Bioanal Chem 2014;406:5989-95.
Stokowski LA. Fundamentals of phototherapy for neonatal jaundice. Adv Neonatal Care 2011;11 5 Suppl:S10-21.
Department of Surgery, Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla 90110
Source of Support: None, Conflict of Interest: None
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6]
| Article Access Statistics|
| Viewed||2281 |
| Printed||54 |
| Emailed||0 |
| PDF Downloaded||24 |
| Comments ||[Add] |