Mobile Phone-Based Detection of Neonatal Jaundice Oral Presentation #1 MobileMed Enterprises Christina Baker Giselle Fontela Pierce Jones Brendan Lynch Sloan Sypher
Problem Statement A non-invasive, cost-effective tool for measuring transcutaneous bilirubin levels has not yet been developed, indicating there is no means to accurately and efficiently diagnose neonatal jaundice in resource-constrained settings.
Background Neonatal jaundice affects ~60% of all newborns Pathological past first 14 days of life Simple to treat with early diagnosis Current methods of diagnosis in Sub-Saharan Africa highly indeterminate
Objective To develop a smartphone application that detects neonatal jaundice by measuring skin reflectance at specific bilirubin-associated wavelengths through the RGB specifications of the phones camera when held up to an infants forehead.
Performance Criteria Constraints Specificity and sensitivity of detection Portable Real time quantification (ideal) Offline vs online cost Limitations Fully-functional and field-ready app before April 19 Access hardware SMS vs MMS, phone service Exclusions Infants between 1-14 days
Solution Characterize response curve Gather clinical data with/without filter Correlate response to gold standard Filter? Design phone case Write software to analyze data. Internally or in the cloud? Troubleshoot Test final phone app in clinical setting Adjust response curve with algorithm
Goals Short-term (1-2 weeks) Correct response curve against background/noise Develop algorithm to adjust response curve Get approval for clinical testing Long-term (>2 weeks) Compile research for Weiner Matrix Obtain clinical data Define parameters for computer scientists
Factors to Consider What did we think about – process On skin - background Cannot alter phones internal settings Necessity of additional hardware? External server necessary for data processing? Simple user interface
Performance Metrics Smartphone App Test Smartphone$299 Lens~$300 Total$599 Blood Serum Test Device$2000-$4000 Disposable Probe $5/test Total$2000-$4000
Experiment: Characterization of Response Curve
Methods: Characterization of Response Curve Access the smartphone camera remotely Spectrometer provides light at 5 nm increments from 350 nm to 800 nm Capture images at each of these wavelengths Analyze intensity of light at each wavelength using MatLab code
Experiment: Characterization of Response Curve
Setbacks Saturated light Neutral density filter Optimizing MatLab code
Conclusions Discovered overlap between RGB curves
Future Considerations RGB response curves adequate to map to full white light spectrum Background light noise from spectrometer must be removed from images Correction needed for one RGB level versus another