Skip to content

and-per-i/NeuroMetric

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

46 Commits
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 

Repository files navigation

๐Ÿง  NeuroMetric V3.2: Kinetic Video Analysis for Neurological Diagnosis

"Markerless" Computer Vision System based on Deep Learning for the quantification and classification of facial motor anomalies.

Python PyTorch MediaPipe License

๐Ÿ“‹ Overview

NeuroMetric is an Artificial Intelligence framework developed to support the objective diagnosis of Movement Disorders starting from simple RGB videos, without the use of wearable sensors.

The system uses a hybrid pipeline combining geometric feature extraction via MediaPipe with deep temporal analysis via recurrent neural networks (LSTM). The current version (V3.2) addresses the critical issues of previous models by introducing a Hybrid Vector (Local + Global) architecture capable of distinguishing micro-movements (tremors) from macro-movements (complex tics/neck jerks).

๐Ÿš€ Key Features (V3.2)

  • ๐ŸŽฅ "In the Wild" Markerless Analysis: Works on standard videos with unconstrained subjects, thanks to a robust Kinematic Normalization pipeline (Nose Anchoring + Scale Invariance).
  • ๐Ÿง  Hybrid Architecture & "The Boost": Combines 248 local features (eyes/mouth) with global head velocity vectors amplified (Feature Scaling x100) to detect complex tics and Head Jerks otherwise invisible.
  • โšก Physiological Data Augmentation: Training on synthetic datasets generated by simulating the physical laws of symptoms (Harmonic oscillators for tremor, Impulse functions for Tics, Random Walk for Dyskinesia).
  • ๐Ÿ“Š Interactive Clinical Reporting: Automatically generates an HTML dashboard with statistics and event timelines of detected anomalies in a loop for medical analysis.
  • ๐Ÿ›ก๏ธ Temporal Stabilization: Post-processing algorithms (Patience & Visual Hold) to eliminate flickering and provide stable and readable detections.

๐Ÿฉบ Supported Pathologies

The model classifies 6 distinct states based on movement kinematics:

  1. ๐ŸŸข NORMAL: Physiological movement.
  2. ๐ŸŸ  TIC (Tourette Syndrome): Ballistic, rapid, and jerky movements (including neck jerks).
  3. ๐Ÿ”ด TREMOR (Parkinson/Essential): Rhythmic oscillations.
  4. ๐ŸŸฃ HYPOMIMIA (Depression/Parkinson): Pathological reduction of expressivity and motor gain.
  5. โšซ PARESIS (Bell's Palsy): Static and dynamic asymmetry (Droop).
  6. ๐Ÿ”ต DYSKINESIA (Tardive/Chorea): Slow, fluid, and chaotic involuntary movements.

About

Deep Learning framework for kinetic analysis of neurological symptoms using LSTM

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages