Early detection for the brain.

NeuroShield is non-invasive, real-time screening of Schizophrenia and Alzheimer’s Disease from EEG signals. A pocketable ESP32 + BioAmp headset, a Conv1D-LSTM model at 94.2% accuracy, and a mobile-first clinical workflow. Screening in minutes, not weeks.

Request access How it works

What you get

Multi-output analysis: simultaneous Schizophrenia + Alzheimer’s risk scores (0–1) from a single 30 s EEG window via a shared CNN-LSTM
Clinical biomarkers: delta / theta / alpha / beta / gamma band powers, spectral entropy, Hjorth parameters, peak alpha frequency
Portable hardware: ESP32 + BioAmp EXG Pill, dry electrodes, 256 Hz sampling, BLE streaming, and setup in under 2 minutes
Mobile-first workflow: React Native app in apps/mobile/ with BLE pairing, voice-assisted recording, and patient / caregiver / clinician roles
Real-time processing: continuous streaming with instant risk scoring, with full reports in under 3 minutes
Privacy-first architecture: on-device processing with end-to-end encryption; data never leaves the device unless you allow it
Full-stack & open: backend in apps/backend/, web in apps/web/, firmware in firmware/, training in ml/

How it works

Three steps from headset to risk score.

Connect: pair the ESP32 + BioAmp over BLE. Dry electrodes, no gel. Place and go.
Record: single-channel EEG at 256 Hz with live signal-quality feedback until ~30 s of clean trace.
Read the scores: the Conv1D-LSTM + attention network returns two calibrated 0–1 risks: schizophrenia 0.12 · alzheimer’s 0.08 in the example above.

Docs

Quickstart · guide · install: read
Hardware · guide · hardware: read
Technology · guide · model: read
How it works · guide · workflow: read

Quickstart

# 1 · pair the headset over BLE $ pair → ESP32 + BioAmp → connected # 2 · record 30 s at 256 Hz in the app $ record --channel 1 --rate 256 # 3 · read the scores (<3 min) $ schizophrenia 0.12 · alzheimer's 0.08 · low risk

Benchmarks

256 Hz sampling, 94.2% detection accuracy, <3 min analysis, 2 conditions screened at once
Model: Conv1D(1→32→64→128) trunk + 2-layer BiLSTM-64 + attention, dual 0–1 risk heads
Trained on: PhysioNet EEG MMI, TUH EEG Corpus, DEED, ADNI, COBRA
Use cases: clinical screening, research studies, remote monitoring, resource-limited settings

FAQ

What conditions can NeuroShield detect?

Schizophrenia and Alzheimer’s Disease: simultaneous 0–1 risk scores from one recording via a multi-output network.

How accurate is the AI analysis?

94.2% on validated sets (PhysioNet MMI, TUH, DEED, ADNI, COBRA) with a shared Conv1D-LSTM + attention trunk.

Is the hardware difficult to set up?

No. Dry electrodes, no gel, BLE pairing. Under 2 minutes from box to recording.

Is patient data secure?

On-device processing, end-to-end encryption, explicit consent before anything leaves the device.

Can this work in resource-limited settings?

That’s the point: low-cost portable hardware + AI analysis delivers specialist-level screening with no neurologist on-site.