All projects
AI · Digital Health2026Completed

NeuroScan

A 16-second webcam test for early stroke signs

An in-browser stroke screening tool that detects facial drooping from 468 facial landmarks — with no data ever leaving the device.

Presented at
IEEE ICADHI 2026 — International Congress on AI & Digital Health Innovation
Try the demo GitHub

The demo's backend is currently offline, so some features may not work.

The problem

Stroke is a race against time, and facial drooping is one of the earliest visible signs in the FAST protocol. But most people can't tell whether a face is drooping — and screening tools that upload video raise real privacy concerns.

The approach

  1. 1Used Google MediaPipe's 468 facial landmarks to run a five-phase guided test — rest, smile, close eyes, open wide, smile again — mirroring what emergency doctors look for.
  2. 2Measured how each side of the face responds to a command, not just how it looks at rest, so the test stays robust even when asymmetry already exists at baseline.
  3. 3Ran everything in the browser, so no video or image ever leaves the user's device.
  4. 4Added a second mode, Photo Scan: a trained machine-learning model analyses facial asymmetry from a single uploaded image, for people who can't complete a live test.

Outcome

  • Presented at IEEE ICADHI 2026, Daffodil International University
  • Two screening modes: a 16-second live test and a single-photo scan
  • Privacy by design — live analysis runs entirely on-device

Built with

MediaPipe Face MeshComputer VisionMachine LearningJavaScriptPythonVercel