AbhishekGowdaL05/AI-Based-Parkinson-Disease-Monitoring-and-Progression-Prediction

🧠 Parkinson Disease Monitoring System

AI-Based Parkinson Disease Progression Prediction and Risk Assessment using Wearable Biomarkers, LSTM Deep Learning, and Machine Learning

📌 Overview

Parkinson’s disease is a progressive neurological disorder that gradually affects:

• —motor control
• —tremor behaviour
• —gait stability
• —movement consistency
• —sleep quality
• —physical activity

Traditional healthcare often detects worsening symptoms only during occasional hospital visits, which may delay treatment.

This project introduces an AI-powered continuous monitoring system that uses wearable biomarker data to:

✅ Predict how much Parkinson’s disease has progressed ✅ Classify the patient’s current risk level ✅ Generate personalized medical recommendations ✅ Create a final AI-generated health report

🎯 Problem Statement

Parkinson’s disease progression is gradual.

A patient may show:

• —worsening tremors
• —reduced gait stability
• —declining activity
• —sleep disturbances
• —inconsistent movement

These changes happen over weeks and months, but hospitals usually see only occasional snapshots.

Our Solution

Use wearable biomarker data + AI to:

• —monitor patients continuously
• —detect progression trends
• —predict severity early
• —assist doctors and caregivers

🏗 Project Architecture

Wearable Biomarker Data
        ↓
Synthetic Dataset Generator
        ↓
LSTM Model (Progression Prediction)
        ↓
Risk Classifier (LOW / MEDIUM / HIGH / CRITICAL)
        ↓
Recommendation Engine
        ↓
AI Medical Report

📊 Input Biomarkers

Our system uses 8 wearable biomarker parameters

🎯 Outputs

The system predicts:

1. Progression Score (0–100)

A continuous severity score indicating:

2. Risk Level

Categorical output:

• —LOW
• —MEDIUM
• —HIGH
• —CRITICAL

3. Medical Recommendations

Examples:

• —Specialist consultation
• —Fall-risk prevention
• —Daily monitoring
• —Medication review
• —Caregiver support

🤖 Deep Learning Model — LSTM

Why LSTM?

Parkinson progression is time-series behaviour

Not:

today only

But:

Week 1 → Week 2 → Week 3 → Week 4 ...

LSTM learns:

• —trend progression
• —temporal deterioration
• —symptom worsening patterns

LSTM Architecture

Input Layer (10 weeks × 8 biomarkers)
        ↓
2-Layer LSTM
        ↓
Dropout
        ↓
Dense Layer
        ↓
Progression Score

Purpose of each layer

Input Layer

Receives:

10 previous weeks of biomarker history

LSTM Layer

Learns:

• —memory of past symptoms
• —temporal progression
• —disease worsening trends

Dropout Layer

Prevents:

• —overfitting
• —memorization

Dense Layer

Converts learned temporal patterns into:

Progression Score

🌲 Machine Learning Risk Classifier

We use:

Random Forest Classifier

Purpose:

Converts:

Progression biomarker state

into:

LOW / MEDIUM / HIGH / CRITICAL

📈 Model Performance

LSTM

• —Stable convergence
• —Predicts progression score using sequence learning

Risk Classifier

Achieved:

Accuracy ≈ 78%

This is realistic because healthcare biomarker classes overlap.

🗂 Folder Structure

pd-monitor/
│
├── data/
│   ├── synthetic_training_data.csv
│   ├── synthetic_test_data.csv
│   └── generate_data.py
│
├── models/
│   ├── lstm_progression_model.pth
│   ├── risk_classifier.pkl
│   ├── scaler.pkl
│   ├── label_encoder.pkl
│   ├── lstm_model.py
│   └── train.py
│
├── utils/
│   ├── predictor.py
│   ├── report_generator.py
│   ├── helpers.py
│   └── constants.py
│
├── reports/
│   └── generated_reports/
│
├── app.py
├── requirements.txt
├── README.md
└── main.py

⚙ Installation

Clone repo

git clone https://github.com/your-username/pd-monitor.git
cd pd-monitor

Create virtual environment

python -m venv .venv
source .venv/bin/activate

Install dependencies

pip install -r requirements.txt

🚀 How to Run

1. Generate synthetic data

python data/generate_data.py

2. Train models

python models/train.py

3. Test predictor

python utils/predictor.py

4. Launch dashboard

streamlit run app.py

🧪 Example Prediction

Input

Heart Rate            : 92
SpO2                  : 94
Temperature           : 36.8
Activity Intensity    : 1.8
Movement Consistency  : 4.5
Tremor Severity       : 7.2
Gait Stability        : 3.4
Sleep Quality         : 4.2

Output

Predicted Progression Score : 65.54
Predicted Risk Level       : HIGH

Recommendations

• —Immediate specialist consultation advised
• —Daily wearable monitoring required
• —Medication review recommended
• —Fall-risk prevention measures required

🌍 Real-World Impact

This project helps:

Patients

Early disease monitoring

Families

Continuous care support

Doctors

Objective progression analysis

Hospitals

AI-assisted clinical decision support

🔮 Future Improvements

• —Real wearable device integration
• —Real patient hospital datasets
• —Fall detection module
• —Caregiver alert system
• —Long-term personalized treatment forecasting

🛠 Tech Stack

• —Python
• —PyTorch
• —Scikit-learn
• —Pandas
• —NumPy
• —Streamlit
• —Joblib

Final Summary

This project uses wearable biomarker time-series data and deep learning to continuously monitor Parkinson patients, predict disease progression, classify severity, and generate AI-powered medical recommendations before the condition worsens.