Smartphone Nailbed Anemia Screening

ISEF Category: Translational Medical Science

Subcategory: Disease Detection and Diagnosis  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A phone camera can catch tiny color changes your eye misses. That matters when a faint pink nail bed can hint at blood flow, oxygen delivery, or low hemoglobin. If you can measure those changes well, you can turn a casual photo into a screening tool. That is the kind of project that feels simple on the surface, then gets deep fast.

What Is It?

This project asks whether standardized smartphone photos of the fingernail bed can help screen for iron-deficiency anemia. You are not diagnosing anyone. You are testing whether color and capillary refill patterns, the time and color change after pressure is released, carry a measurable signal that matches anemia status.

Think of it like calibrating a bathroom scale. A raw photo is not enough, because light, skin tone, camera settings, and angle all change the result. Your job is to control those inputs, measure the image features that matter, and see whether the pattern stays useful across a diverse set of skin tones. Fitzpatrick stratification means grouping by skin tone type so you can test whether the model works fairly across groups.

Why This Is a Good Topic

This is a strong science fair topic because you can ask a clear question, collect original data, and test real-world usefulness. It connects to anemia screening, which matters in clinics, schools, and places with limited medical access. You can learn image standardization, data labeling, statistics, and basic machine learning, all from one project.

Research Questions

• How does standardized lighting change the accuracy of nail-bed color measurements for anemia screening?
• What is the effect of Fitzpatrick skin-tone group on the model's sensitivity and specificity?
• Does adding capillary refill timing improve prediction more than color alone?
• To what extent do white-balance correction and color calibration reduce between-phone variation?
• Which image features, such as red-to-green ratio or brightness, best separate low-hemoglobin and normal samples?
• How does a model trained on a diverse dataset perform on a held-out skin-tone group?
• What is the effect of different nail regions, such as the nail bed versus the cuticle area, on screening performance?

Basic Materials

• Smartphone camera with manual control settings.
• Neutral background and consistent light source.
• Color calibration card or printed color reference.
• Tripod or phone stand.
• Ruler or fixed-distance spacer.
• Consent forms and participant information sheet.
• Spreadsheet software for data logging.
• ImageJ for image measurements.
• Notebook for metadata and session notes.

Advanced Materials

• Digital colorimeter or spectrophotometer.
• Clinical hemoglobin reference data, if available through an approved partner.
• Diffuse light box or controlled imaging enclosure.
• Cross-polarizing filters for the camera and light source.
• Reference skin-tone or color calibration standards.
• Computer with Python, OpenCV, and scikit-learn.
• Access to a supervised clinical or university research setting for sample collection.
• Secure database for de-identified participant records.

Software & Tools

• ImageJ: Measures color channels and region-of-interest values from nail images.
• Python: Cleans data, builds features, and runs prediction models.
• OpenCV: Automates image standardization and color correction.
• scikit-learn: Trains baseline classifiers and tests model performance across groups.
• Excel or Google Sheets: Organizes metadata, labels, and summary statistics.

Experiment Steps

1. Define the exact screening question you want to test, including the outcome label and the participant groups you will compare.
2. Choose one image standardization plan, then lock your camera, lighting, background, and distance rules before collecting data.
3. Set up a labeling scheme for nail-bed region, capillary refill timing, and skin-tone group so every photo gets the same treatment.
4. Build a feature list that turns each image into numbers, then decide which features will form your baseline model.
5. Plan controls that separate lighting effects, skin-tone effects, and phone-camera effects so you can test fairness and generalization.
6. Predefine your evaluation metrics, including sensitivity, specificity, and subgroup performance, before you train the model.

Common Pitfalls

• Letting room light change between sessions, which shifts nail color values more than the biology does.
• Mixing photos from different phone models without correction, which teaches the model camera differences instead of anemia signals.
• Using skin tone labels without enough samples in each group, which makes subgroup results unstable.
• Cropping the nail region by hand with inconsistent rules, which adds measurement noise from frame to frame.
• Treating capillary refill as a yes-or-no label without a consistent timing method, which blurs the signal you are trying to model.

What Makes This Competitive

A class-level version of this project only reports whether photos and anemia labels match. A stronger version tests fairness across skin-tone groups, compares multiple feature sets, and checks whether a model still works on new phones or new lighting setups. You can also raise the bar by using a blinded labeling plan and a proper held-out test set. That turns the project from a simple image exercise into a real study of clinical screening reliability.

Project Variations

• Compare nail-bed color features with lower-eyelid color features for anemia screening.
• Test whether capillary refill timing adds more value than color features alone across different skin tones.
• Train and evaluate the model on images from two phone brands to measure camera-domain shift.

Learn More

• PubMed: Search for review articles on anemia screening, capillary refill, and smartphone-based medical imaging.
• NIH National Library of Medicine: Look for patient education and clinical background on iron-deficiency anemia.
• NIH PubMed Central: Read free full-text papers on image-based screening and skin-tone bias in medical AI.
• ImageJ Documentation: Find tutorials for measuring color channels and regions of interest in images.
• MIT OpenCourseWare, Introduction to Machine Learning: Use free lecture materials to understand training, validation, and error metrics.
• NIH Clinical Center: Search for research ethics and informed consent guidance for human-subject projects.