Skip to content
All work

Research, 2026

Schamroth window detection

Computer vision that screens for finger clubbing, a sign of lung and heart disease. Published in IEEE.

Role
Research project, published in IEEE
Type
Research
Links
Source on GitHubRead the paper
published 2026
IEEE
clinical risk levels
5
image test set
65

The problem

Clubbed fingers can be an early sign of lung cancer and heart disease. Doctors check with the Schamroth window test: two fingers back to back, looking for a diamond-shaped gap of light. Image analysis can make that screening easier to access.

How I approached it

  1. 1

    Find the fingers

    MediaPipe hand landmarks locate the opposing index fingertips, with a centre-based fallback for close-up frames and image-quality gates before analysis.

  2. 2

    Measure the gap

    A gap-presence score built from bright/dark pixel ratios, contrast and left-right symmetry, after CLAHE enhancement.

  3. 3

    Evaluate honestly

    An evaluation harness reporting accuracy, sensitivity, specificity, F1 and a confusion matrix over 65 images of healthy, simulated and clinical cases.

What I built

  • The score maps to 5 clinical risk levels with quality-weighted confidence, wrapped in a Streamlit app.

The result

Published as 'Early Lung Cancer and Heart Disease Prediction from Clubbed Fingers Using Machine Vision' (IEEE, 2026).

Built with

  • Python
  • MediaPipe
  • OpenCV
  • Streamlit