A real-time interactive platform for visualizing Parkinson's disease motor symptoms through anatomically-accurate motion silhouettes and multi-modal data analysis.
Web Application: https://parkinson-viz.onrender.com/
Performance Note: The cloud deployment has animation limitations due to browser ↔ server latency. For optimal 60fps real-time motion visualization, local deployment is recommended.
This system transforms clinical Parkinson's disease data into intuitive, real-time anatomical motion visualizations. It integrates multiple data modalities to provide comprehensive patient assessment through:
- Real-time motion silhouettes with anatomically-accurate human figures
- Multi-modal data integration (gait, clinical scores, digital sensors)
- Interactive analysis tools for bilateral movement comparison
- Clinical feature engineering with evidence-based composite indices
- 60fps smooth animation (local deployment)
Anatomically-Accurate Silhouettes: 8-head proportional human figures with proper joint articulation
Patient-Specific Animation: Motion patterns derived from individual gait measurements
Multi-Modal Analysis: Gait data + UPDRS scores + Digital sensors + Demographics
Real-Time Controls: Play/Pause/Speed controls with phase-locked motion
Clinical Insights: Bilateral asymmetry, movement quality, severity staging
Interactive Visualizations: Correlation plots, radar charts, motion cycles
Raw Data Sources → Data Integration → Feature Engineering → Real-Time Visualization
↓ ↓ ↓ ↓
- Gait Data → Multi-Dataset → Clinical Scores → Motion Silhouettes
- UPDRS Scores → Merge & Clean → Motion Features → Interactive Plots
- Digital Sensors → Quality Control → Composite Indices → Animation Controls
- Demographics → Missing Data → Bilateral Analysis → Performance Metrics
- Frontend: Dash (Python) + Plotly.js for interactive visualization
- Backend: Python with Pandas, NumPy, SciPy for data processing
- Animation: Custom SVG-based silhouette generator with real-time updates
- Data Processing: Multi-threaded callback system for smooth animations
- Deployment: Render.com cloud platform + local deployment support
The system uses comprehensive multi-modal datasets from Parkinson's disease research:
-
Motor Assessments (192 patients): Gait analysis with 60+ motion parameters
- Arm swing amplitudes (LA_AMP_U, RA_AMP_U)
- Gait speed and cadence (SP_U, CAD_U)
- Movement asymmetry (ASA_U) and smoothness (JERK)
- Timed Up & Go (TUG) test measurements
-
UPDRS Clinical Scores: Standardized severity assessments
- Part II: Patient-reported motor symptoms
- Part III: Clinician-assessed motor examination
- Part IV: Motor complications
-
Digital Sensor Data (108,901 measurements): High-frequency behavioral signals
- Drawing tests (spirals, lines, circles)
- Voice analysis and finger tapping
- Daily activity monitoring
-
Demographics & Status: Patient stratification variables
- Age, sex, handedness, race/ethnicity
- Cohort definitions and enrollment status
- PD Patients: Primary Parkinson's disease cohort
- Healthy Controls: Neurologically normal comparison group
- Prodromal: Early-stage/at-risk individuals
- SWEDD: Subjects Without Evidence of Dopaminergic Deficit
# Python 3.8 or higher
python --version
# Required packages
pip install dash plotly pandas numpy scipy- Clone the Repository
git clone https://github.com/erickim73/parkinson-viz.git
cd parkinson-viz- Create Virtual Environment (Recommended)
python -m venv venv
# Windows
venv\Scripts\activate
# macOS/Linux
source venv/bin/activate- Install Dependencies
pip install -r requirements.txt- Prepare Data Directory Structure
parkinson-motion-viz/
├── Curated_Data_Cuts/
| ├── PPMI_Curated_Data_Cut_Public_20241211.xlsx
├──Medical_History
| ├──Features_of_Parkinsonism_06_Jan2025.csv
| ├──Neuological_Exam_05Jan2025.csv
| ├──Other_Clinical_Features_06Jan2025.csv
├── Motor_Assessments/
│ ├── Gait_Data___Arm_swing_06Jan2025.csv
| ├──Gait_Substudy_Gait_Mobility_Assessment_and_Measurement_06Jan2025.csv
│ ├── MDS_UPDRS_Part_II__Patient_Questionnaire_06Jan2025.csv
│ ├── MDS-UPDRS_Part_III_06Jan2025.csv
│ └── MDS-UPDRS_Part_IV__Motor_Complications_06Jan2025.csv
├── Study_Docs/
| ├──Code_List_-__Annotated__08Jan2025.csv
| ├──Data_Dictionary_-__Annotated__08Jan2025.csv
├── Subject_Characteristics/
│ ├── Demographics_08Jan2025.csv
│ └── Participant_Status_08Jan2025.csv
└── app.py
- Run the Application
python app.py- Access the Application
Open your browser and navigate to:
http://127.0.0.1:8050/
Loading datasets...
✓ Loaded gait data: (192, 60)
✓ Loaded UPDRS III: (xxx, xx)
✓ Loaded UPDRS II: (xxx, xx)
✓ Loaded demographics: (xxx, xx)
✓ Loaded status: (xxx, xx)
✓ Final merged dataset: (xxx, xx)
✓ Available patients: xxx
Starting Enhanced Multi-Modal Parkinson's Dashboard...
Dash is running on http://127.0.0.1:8050/
- Choose a patient from the dropdown menu
- The system loads patient-specific gait and clinical data
- Motion silhouette updates to reflect individual movement patterns
- Gait/Walking Test: Continuous walking motion with arm swing
- TUG Test: Timed Up & Go sequence (sit→stand→walk→turn→sit)
- Postural Sway/Balance: Standing balance with center-of-mass movement
- Free Motion: Unconstrained movement patterns
- Play/Pause: Control real-time motion animation
- Speed: Adjust animation speed (0.1x to 3.0x)
- Reset: Return to initial animation phase
- Correlation Analysis: X/Y axis feature selection for scatter plots
- Bilateral Comparison: Left vs. right arm movement analysis
- Motion Quality: Radar chart of movement characteristics
- Gait Cycle: Phase-locked motion patterns over time
The core innovation is the anatomically-accurate motion silhouette system:
class MotionSilhouetteGenerator:
"""Anatomically-correct motion visualization"""
def create_anatomical_silhouette(self):
# 8-head figure proportions with realistic body segments
return {
'head': {...}, # Proportional head structure
'torso': {...}, # Central body reference
'left_upper_arm': {...}, # Bilateral arm segments
'right_upper_arm': {...},
'left_forearm': {...},
'right_forearm': {...},
'left_hand': {...},
'right_hand': {...},
'left_thigh': {...}, # Leg segments with joints
'right_thigh': {...},
# ... additional body parts
}The system uses a sophisticated 4-callback architecture for smooth 60fps animation:
- Patient Data Store (Low Frequency): Caches patient data to avoid expensive lookups
- Animation Control (User Triggered): Manages play/pause/speed controls
- Time Phase Ticker (High Frequency - 100ms): Lightweight phase incrementation
- Silhouette Renderer (Data Driven): Main rendering engine for motion visualization
Patient-specific motion is calculated using clinical measurements:
def calculate_gait_motion(left_arm_amp, right_arm_amp, gait_speed, time_phase, asymmetry):
# Normalize amplitudes to realistic range
left_swing = (left_arm_amp / 50.0) * math.sin(time_phase) * 0.5
right_swing = (right_arm_amp / 50.0) * math.sin(time_phase + math.pi) * 0.5
# Apply clinical asymmetry
asymmetry_factor = min(asymmetry / 10.0, 0.3)
left_swing *= (1 + asymmetry_factor)
right_swing *= (1 - asymmetry_factor)
# Synchronize leg motion
speed_factor = min(gait_speed, 1.5)
leg_phase = time_phase * speed_factor
left_leg_swing = math.sin(leg_phase + math.pi) * 0.3
right_leg_swing = math.sin(leg_phase) * 0.3| Aspect | Local Deployment | Cloud (Render.com) |
|---|---|---|
| Animation FPS | 60fps (smooth) | ~10fps (laggy) |
| Data Processing | Direct memory access | Browser ↔ Server latency |
| User Experience | Optimal | Limited |
| Setup Required | Python environment | None |
| Cost | Free | Free tier (with limitations) |
The cloud deployment suffers from a fundamental architectural challenge:
- Issue: Real-time animation data travels browser → server → browser on each frame
- Impact: Noticeable lag, reduced frame rate, poor user experience
- Solution: Local deployment eliminates network latency entirely
# Efficient data handling for large datasets
def enhance_merged_data(self, merged):
"""Optimized feature engineering"""
# Select only essential columns to minimize memory usage
essential_cols = ['PATNO', 'ASA_U', 'SP_U', 'RA_AMP_U', 'LA_AMP_U']
# Vectorized operations for speed
merged['ARM_ASYMMETRY'] = abs(merged['RA_AMP_U'] - merged['LA_AMP_U']) / \
(merged['RA_AMP_U'] + merged['LA_AMP_U'] + 1e-6)- Movement Biomarker Discovery: Novel patterns in motion data
- Treatment Response Monitoring: Visualize therapy effectiveness
- Disease Progression Tracking: Longitudinal movement changes
- Cohort Comparison Studies: Control vs. patient populations
- Objective Assessment: Transform clinical scores into visual patterns
- Patient Education: Help patients understand their movement patterns
- Therapy Planning: Identify specific movement impairments
- Progress Monitoring: Track improvement over time
- Medical Training: Visualize Parkinson's motor symptoms
- Patient Understanding: Intuitive representation of clinical data
- Research Presentation: Compelling visualization of findings
Immediate Improvements:
- Client-side animation rendering for cloud deployment
- Progressive data loading for faster initialization
- Enhanced mobile responsiveness
- Export functionality for motion videos
Advanced Features:
- 3D motion visualization with depth perception
- Machine learning severity prediction from motion patterns
- Multi-patient comparison views
- Real-time sensor data integration
Clinical Integration:
- DICOM compatibility for medical imaging integration
- HL7 FHIR standards for clinical data exchange
- Automated report generation
- Clinical decision support recommendations