Mobile Phone Chromatic Monitoring of Jaundice In Vivo
A. T. Sufian, C. R. Jones and H. M. Shabeer
The occurrence of jaundice (bilirubin) in newborn babies can lead to serious mental deficiencies in the baby. The established procedure for diagnosing bilirubin is by extracting and analysing a blood sample from a baby (invasive blood sampling). A possible alternative to such invasive blood sampling is the use of transcutaneous bilirubinometry, which involves optically monitoring the skin tissue to avoid the need for disruptive blood extraction. The first transcutaneous bilirubinometer was introduced in 1980 (Yamamanouchi et al„ 1980). Since then, several other devices have been developed, and important adjustments (such as the correction for the presence of other skin chromophores [e.g., melanin, haemoglobin]) have been made to improve their accuracy. These second-generation bilirubinometers are suitable for the screening of hyperbilirubinaemia, leading to a considerable decrease of the number of hospital re-admissions. However, after more than 30 years of development, no transcutaneous bilirubinometer has proven itself a worthy replacement for invasive blood sampling. Reasons for this limited clinical value are diverse (e.g., the technological design of the bilirubinometers, the method of clinical evaluation and variations between patients) but have not been investigated thoroughly in the literature.
A chromatically based approach has therefore been investigated in a new form of bilirubinometer. The approach is based upon a light-tight, handheld unit incorporating a mobile phone, an optical referencing template and sample window illuminated with white-light light-emitting diodes (LEDs) (Sufian et al., 2018). Various optical chromatic corrections are conveniently incorporated, making the unit suitable for use in remote locations which have mobile phone system access.
Chromatic Monitoring Unit
The monitoring unit is based upon chromatically addressing white light from a group of light- emitting diodes reflected/scattered from the skin tissue of a neonate in vivo plus optical references using a mobile phone camera. Figure 6.1 shows such a monitoring unit. Figure 6.2 shows a schematic diagram of the overall system.
The system consists of a removable mobile phone camera mounted upon a light-tight housing for monitoring a reference template carrying a rigid plastic window at the other end of the light-tight enclosure. The rigid plastic window needs to be pressed against the skin tissue under test (e.g.,
FIGURE 6.1 Light-tight handheld unit with mobile phone.
FIGURE 6.2 Schematic diagram of the handheld chromatic unit.
FIGURE 6.3 Template images captured by a mobile phone camera [WO (set B1 = 0). W2(set B1 = 10 mg/dL) and WT (neonate nose tissue) in the sample window set reference ro = (set Bl « 0), r2 = (set B1« 10)].
the nose) in order to occlude blood vessels from the tissue and so enhance the optical influence of the tissue. The template, including the plastic window, is illuminated by an array of white-light surface-mounted diodes arranged in a strip (0.2 W) and circled around an aperture (Figure 6.2), through which the camera lens observes and captures an image of the template and window (Sufian et al„ 2018).
The template consists of a colour-printed paper with a rectangular and rigid sample viewing plastic window (Figure 6.2). The printed paper has several sectors with different chromatic signatures (Figure 6.3), which are empirically chosen, along with some alignment-assisting features. Tw'o template sectors adjacent to the plastic window have chromatic signatures chosen empirically from occluded skin images of two neonates, one without jaundice (ro) and one with 10 mg/dL bilirubin (r2). These areas provide an indication relative to a sample in the window as to w'hether the camera settings and so on vary so that compensation can be made. The remaining template areas are empirically coded to accommodate automatic image adjustments made by the camera operating system and can vary for different cameras. If the camera is operated in “auto mode”, the exposure time is automatically set. Geometric alignment features on the template (Figure 6.3) (red dots for correct angular orientation, square black frame for image size) ensure that the camera automatically addresses the relevant template sectors (ro, r2, W) for chromatic analysis.
Primary Chromatic Processing
Relevant areas of sectors on the image captured by the mobile phone camera (sample window [WT], reference sectors [ro, r2]; Figure 6.3) produce outputs for the three chromatic parameters R. G, B. The wavelength responses of R, G, В are non-orthogonal, as shown in Figure 6.4. They cover the wavelength ranges of various skin tissue components (bilirubin, haemoglobin, melanin; Bhutani et al„ 2000), as indicated in Figure 6.4. Thus, melanin (Me) and haemoglobin (He) are covered by all three processors (R. G, B), whilst bilirubin is covered mainly by processor В and to some extent by G.
The R, G, В outputs may be processed to yield chromatic parameters X, Y, Z and L (Chapter l), with X, Y, Z being represented on a two-dimensional chromatic map (Chapter l). Some preliminary chromatic signatures of 48 different in vivo tissue samples w'ith different levels of bilirubin obtained with several different mobile phone cameras are shown on the Y:Z map of Figure 6.5. This illustrates the complex distribution of data produced by such basic chromatic processing.
FIGURE 6.4 Typical overlapping wavelength responses (R. G. B) of an electronic camera relative to wavelength ranges of some skin tissue components (bilirubin [Bl], haemoglobin [He], melanin [Me]) (Sufian et al„ 2018).
FIGURE 6.5 Basic Y:Z chromatic map of bilirubin test results
FIGURE 6.6 Primary chromatic bilirubin PCh(BL) versus total serum bilirubin [TSB (mg/dL)] (48 neonate test results, six different phone cameras; solid line —> regression line, dashed lines —> 95% prediction interval boundaries) (Sufian et al., 2018).
A basic chromatic calibration curve for bilirubin levels may be derived from the chromatic map of Figure 6.5 using a calibration parameter (Y/Z), which is a form of di-stimulus dominant wavelength (Sufian et al., 2018) obtained from the gradient of the locus from the origin to the Y, Z point.
A linear variation of this ratio between the set values for B1 = 0 and 10 mg/dL represented by the Y/Z values of reference papers Wo (B1 = 0) and W2 (B1 = 10 mg/dL) in the sample window may be assumed to provide an estimate of the bilirubin level fPCh(Bl)] from the measured value of a tissue (T) in the sample window [(Y/Z)(WT)], that is
Bilirubin levels [PCh(Bl)] obtained with this equation for the 48 in vivo tissue samples with various cameras are shown in Figure 6.6 as a function of B1 values obtained for each neonate from blood tests yielding total serum bilirubin (TSB) levels.
The results have a high degree of scatter, which illustrates the complicating effects of various factors such as the use of different types of cameras, different tissue melanin levels in individual neonates and so on. Consequently, various correction procedures need to be incorporated through the use of secondary chromatic processing.
Secondary Chromatic Processing
Secondary chromatic processing allows corrections to be made for differences between various cameras (e.g., different pixel densities [0.8-8.3 megapixels; Sufian et al.. 2018]; different relative
FIGURE 6.7 Chromatic Y:Z map showing locus of set reference areas (So-S2) and loci of samples in window (Wo-W2) for 12 different cameras (Sufian et al., 2018).
numbers of R, G, В pixels per unit area etc.), template and window distortions and so on. Such effects are compensated via the use of the reference areas on the template (ro, r2) (Figure 6.3) and window references (So, S2) via a Y:Z chromatic compensating map; see Figure 6.7. This shows a locus of the set values So-S2, along with the corresponding loci for each of 12 different cameras.
Such a map is used for producing correction factors (CFs) to account for various complexities by quantifying deviations from the set S0-S2 locus. A correction factor for Z is the ratio of a reference sample in the window (e.g., ZSo) to that of the set reference value (ZWo), that is
A corrected value of a tissue sample [Z(WTo)] is then given by
A similar expression applies with regard to the S2 reference, the choice of reference preferred depending upon the reference nearest to the measured sample value. Furthermore, a similar correction procedure is applied to the chromatic Y parameter.
A calibration graph of Z(WTo) versus the total serum bilirubin using the chromatic test results obtained with the 48 different tissue samples in vivo is shown in Figure 6.8a. Thus, the Z TSB value indicated by the corrected bilirubin level for a tissue sample with Z( WT)* is given by the TSB value ZCh(Bl) indicated by the calibration curve (solid line) (Figure 6.8a).
A corresponding calibration graph of Y(WTo) versus TSB is shown in Figure 6.8b, which likewise yields a corrected bilirubin level of YCh(Bl). The best estimate for a bilirubin level correction is given in practice by
FIGURE 6.8 Calibration graphs for corrected (WT)*: as a function of total blood bilirubin level TSB (mg/ dL). (a) Y(WT)*: TSB. (b) Z(WT)*: TSB (Sufian et al„ 2018).
FIGURE 6.9 Corrected chromatic bilirubin [SCh(Bl)] versus total serum bilirubin (mg/dL) (48 neonate tests, six different phone cameras; solid line —> regression line, dashed lines —>95% prediction interval boundaries) (Sufian et al.. 2018).
Figure 6.9 shows values of the corrected chromatic bilirubin levels for the 48 in vivo tissue samples using the Y(WT)* and Z(WT)* calibration graphs of Figure 6.8. SCh(Bl) as a function of the TSB (mg/dL) was obtained from blood tests.
Compared with total serum bilirubin values (blood tests), the method of correcting the chromatic results has a significantly improved linear regression correlation coefficient (R2) of 0.81 compared with a 0.43 value for the uncorrected chromatic results (Sufian et al., 2018). The chromatic method also compares favourably with test results reported by other investigators, for which R2 « 0.31-0.71 (Sufian et al., 2018). The 95% prediction interval boundaries [PI(95%)] were improved from ±9 mg/dL for the uncorrected results (Figure 6.3) to ±3 mg/dL for the corrected results (Figure 6.9) (Sufian et al., 2018).
Bhutani, V.K. Gourley. G.R., Adler, S„ Kreamer, B., Dalin, C.. and Johnson. L.H. (2000) Noninvasive measurement of total bilirubin in a multiracial predischarge newborn population to assess the risk of severe hyperbilirubinemia. Pediatrics, 106, el7.
Sufian. A.T.. Jones. G.R., Shabeer, H.M. Elzagzoug, E.Y., and Spencer, J.W. (2018) Chromatic techniques for in vivo monitoring jaundice in neonate tissues. Pliysioplogical Measurement, 39(9), 095004. Yamamanouchi, I., Yamanouchi. Y.. and Igarashi, I. (1980) Transcutaneous bilirubinometry: Preliminary studies of noninvasive transcutaneous bilirubin meter in the Okayama National Hospital. Pediatrics, 65, 195-202.
~7 Optical Chromatic