Smartphone Jaundice Screener for Newborns Science Fair

By /

Smartphone Jaundice Screener for Newborns Science Fair

ISEF Category: Biomedical Engineering

Ready to Turn This Idea Into a Real Project?

This guide was put together with the help of AI research tools to give you a solid starting point. But a competitive science fair project lives in the details: refining your research question, fine-tuning your variables, analyzing your data, and presenting your findings like a seasoned scientist.

For next steps tailored to your interests, skill level, and timeline, work one-on-one with a MehtA+ mentor. Learn more about MehtA+ Science & Engineering Research Mentorship →

Subcategory: Biomedical Devices  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

Untreated newborn jaundice can cause permanent brain damage, yet in many parts of the world there is no blood-test lab to confirm it. A smartphone, a 3D-printed color-calibration card, and a CNN can estimate bilirubin from a photo of the sclera. The whole kit costs about 25 dollars and could screen babies in places where bilirubin meters never go.

What Is It?

Bilirubin is a yellow pigment that builds up when a newborn's liver cannot process it fast enough. The yellow tint shows up first in the eyes and then the skin.

A 3D-printed card with known color patches lets the phone calibrate for ambient light. The card sits next to the eye or skin in the photo. Software corrects for the camera's white balance using the patches.

A regression CNN learns the mapping from sclera or skin RGB to bilirubin concentration. Public neonatal jaundice image datasets and published bilirubin values train the model. The output is a continuous estimate, not a class.

Why This Is a Good Topic

Resource-limited neonatal care is a public-health priority. Hardware is cheap and the ML target is a clean regression problem. You will learn color calibration, regression CNNs, and clinical-agreement statistics.

Research Questions

  • How does color-card design change estimation error?
  • What is the effect of ambient lighting on bilirubin prediction?
  • Does sclera or skin RGB give better accuracy?
  • To what extent does skin tone shift bias?
  • Which CNN backbone hits the smallest mean absolute error?
  • How does smartphone model affect estimates?
  • What is the effect of training-set size on error?

Basic Materials

  • Smartphone with rear camera.
  • 3D-printed color-calibration card or printed Pantone-style chart.
  • Standard daylight LED for repeatable lighting.
  • Public neonatal jaundice image datasets (with permissions).
  • Informed-consent template if any new photos are collected.

Advanced Materials

  • Transcutaneous bilirubin meter for ground-truth comparison.
  • Clinical mentor for hospital data access.
  • IRB approval for new imaging studies.
  • Cloud GPU.

Software & Tools

  • PyTorch: Trains the regression CNN.
  • OpenCV: Implements color calibration with the patch card.
  • scikit-learn: Computes calibration plots and Bland-Altman statistics.
  • Roboflow free tier: Manages labeled image datasets.

Experiment Steps

  1. Lock the color-card design and printing process.
  2. Decide whether sclera or skin is your primary region of interest.
  3. Build a subject-wise data split for training and validation.
  4. Calibrate raw RGB to a reference white using the patch card before any model runs.
  5. Train the CNN and produce a calibration plot.
  6. Compare predicted bilirubin to ground-truth meter values across skin tones.

Common Pitfalls

  • Skipping color-card calibration on every image.
  • Training on a narrow skin-tone range and claiming general accuracy.
  • Mixing subjects across train and test.
  • Reporting R-squared without a calibration plot.
  • Ignoring ambient-light differences between hospital and home photos.

What Makes This Competitive

A competitive entry runs strict subject-wise splits, reports calibration plots not just R-squared, and audits fairness across skin tones. Comparing the model against published transcutaneous bilirubin meters and reporting agreement with Bland-Altman analysis is the strongest validation move.

Project Variations

  • Replace sclera RGB with HSV or LAB color spaces and compare.
  • Use the same pipeline for anemia screening from conjunctiva.
  • Add a smartphone flashlight standardization step and remeasure variability.

Learn More

  • PubMed: Search transcutaneous bilirubin smartphone reviews.
  • NIH PubMed Central: Open-access neonatal jaundice papers.
  • American Academy of Pediatrics: Open clinical guidelines on hyperbilirubinemia.
  • OpenCV documentation: Free color-correction tutorials.
  • MIT OpenCourseWare: Course 6.S191 Introduction to Deep Learning.
Shopping Cart