Objective Tremor Measurement from the Phone in Your Pocket
NeuroMotor transforms any LiDAR-equipped smartphone into a clinical-grade neurological assessment device using depth sensing and posture-conditioned frequency analysis.
System Overview
The pipeline takes continuous depth frames and produces a structured assessment report with quantified motor indicators and confidence scores.
Input
Depth sensor captures hand at 30-60 fps
Processing
3D pose estimation & posture classification
Output
Structured assessment with confidence scores
The Core Insight
Different neurological conditions produce tremor in different situations. By measuring tremor separately in each context — what we call posture conditioning — the system can distinguish between conditions that conventional sensors conflate.
Guided Assessment & Compliance Verification
A typical assessment takes 3-5 minutes. The app guides users through motor tasks while continuously verifying compliance.
Animated Instructions
Visual guidance showing target hand posture for each task
Real-Time Verification
Colour-coded overlay confirms correct hand position
Compliance Checking
Validates posture, stability, and motion patterns continuously
Quality Enforcement
Prompts for task repetition if insufficient valid data collected
Why This Matters
In traditional clinical assessments, measurement quality depends entirely on the examiner's skill and patient compliance. Our system enforces standardization automatically, making results comparable across sessions, patients, and clinics.
Data Quality Gating & Confidence Scoring
Multi-criteria validity gates ensure only high-quality data enters the analysis.
Depth Confidence Filter
Monitors per-pixel confidence from LiDAR sensor, flags low-quality depth measurements
Occlusion Detector
Detects when fingers disappear behind other fingers or leave frame
Motion Artifact Classifier
Identifies large involuntary movements that would contaminate frequency analysis
Only data passing all three filters enters frequency analysis. Every output indicator carries a confidence score reflecting data quality and quantity.
Posture-Conditioned Tremor Features
The core innovation: computing tremor features separately for each clinically defined posture state.
Why Posture Conditioning Changes Everything
Consider two patients with a 5 Hz tremor measured by a conventional sensor:
Patient A: Parkinson's
High rest tremor amplitude with low postural amplitude (ratio >> 1)
Patient B: Essential Tremor
Low rest tremor with high postural amplitude (ratio << 1)
Measured Features per Posture State
Peak tremor frequency
Dominant oscillation frequency (3-8 Hz)
Tremor amplitude
Physical fingertip displacement in millimeters
Spectral concentration
Purity of tremor signal vs broadband noise
Harmonic ratio
Second harmonic strength indicating non-sinusoidal patterns
Inter-finger coherence
Synchronization of tremor across fingers
Tremor constancy
Fraction of time tremor is present
Sensor Modality Selection & Normalisation
Works with three depth sensing technologies, adapting automatically to available hardware.
LiDAR (Preferred)
Direct time-of-flight depth measurement with hardware confidence map
Stereoscopic Depth
Depth computed from disparity between dual rear cameras
Monocular Depth
Single camera with ML depth estimation, works on any phone
Cross-Sensor Normalisation
All depth data transforms into a standardized 256×192 format with confidence mapping. This means patients can switch phones between sessions without invalidating longitudinal tracking. Works on ~95% of smartphones sold in the last three years.
Longitudinal Tracking
A single assessment is useful. A trend over time is transformative.
What We Track Over Time
Rate of change: Linear regression slope over time
Variability index: Session-to-session consistency
Change detection: Flags when cumulative change exceeds clinical threshold
Disease Progression
Detect advancement weeks before subjective observation
Treatment Response
Quantify medication effects with precision
Early Detection
Alert when first subtle motor signs appear
Privacy & Deployment
On-Device Processing
The entire pipeline runs on the phone's processor. Total memory footprint under 200 MB. Raw depth imagery processed in volatile memory only—never written to persistent storage.
On-Device Only
Everything stays on the phone
Consumer self-monitoring
Hybrid
Features sync via encrypted transport
Remote patient monitoring
Clinical Integration
HL7 FHIR to EHR systems
Clinical trials & clinics
All modes are GDPR and HIPAA compliant by design