Assessment-of-Parkinsonian-Tremor-Severity-Through-Neurorobotic-Models-and-CNN-based-Classification

Project Overview

This project extends and builds upon previous work (specifically Timothé Petitjean’s thesis) by developing a comprehensive neurorobotics computational framework for Parkinson's Disease (PD). The work combines computational modeling of the Basal Ganglia–Thalamo–Cortical (BG-T-C) loop, sensorimotor integration with the iCub robot simulator, feature extraction from joint movement data, and the application of deep learning models (CNNs) to classify Parkinsonian severity levels.

The project is divided into three main stages:

• Prototype 1: Replication of closed-loop DBS control using the BG-T-C computational model.
• Prototype 2: Sensorimotor integration of the iCub robot with tremor simulation and closed-loop DBS interventions.
• Prototype 3: Feature engineering of robotic kinematic data and CNN-based severity classification.

---

Prerequisites

Before proceeding, please make sure you have installed and compiled the iCub simulation environment and all necessary dependencies from the following repository: 👉 Timothé Petitjean's GitHub Repository

You must follow the setup instructions provided there, including installing:

robotology-superbuild

YARP middleware

Gazebo simulation environment

These components are necessary for the successful execution of the neurorobotics simulations developed in this project.

Docker

To build the Docker container:

cd docker
bash build-docker.sh

If running for the first time, compile the .mod files by running:

nrnivmodl

Computational Model

The computational model needs to be launched outside from the docker.

Download the necessary packages in the requirements.txt beforehand.

To run the computational model:

python MarmosetBG.py

Launch Gazebo Simulation

Install and compile the robotology superbuild before starting simulation.

To launch the Gazebo iCub simulation, first initiate the yarp server:

yarpserver

Then in another Command window launch the gazebo simulation:

gazebo tutorial_joint-interface.sdf

---

Additional Contributions

Sensorimotor Loop Integration

The iCub simulator is connected to the computational model to create a full sensorimotor feedback loop, allowing dynamic modulation of motor behavior based on simulated Parkinsonian neural signals.

Data Extraction

From the iCub simulations, kinematic data including joint angles, joint velocities, tremor amplitude, and frequency are collected to represent motor symptoms.

Feature Engineering

Critical features such as joint velocity, angular acceleration, tremor frequency, and amplitude are extracted from the raw time-series data to create clean machine learning-ready datasets.

MANOVA Statistical Validation

A Multivariate Analysis of Variance (MANOVA) is performed to statistically confirm that the engineered kinematic features significantly differ across different Parkinsonian severity conditions (Healthy, PD without DBS, PD with DBS).

CNN-based Severity Classification

A Convolutional Neural Network (CNN) is designed, trained, and validated to classify Parkinsonian severity levels based on the engineered features. Model evaluation metrics include accuracy, precision, recall, F1-score, and confusion matrices.

---

How to Use

1. Clone the repository:

git clone https://github.com/aatish1741/Assessment-of-Parkinsonian-Tremor-Severity-Through-Neurorobotic-Models-and-CNN-based-Classification.git
cd Assessment-of-Parkinsonian-Tremor-Severity-Through-Neurorobotic-Models-and-CNN-based-Classification

2. Install required Python packages:

pip install -r requirements.txt

3. Run the computational BG-T-C model:

python MarmosetBG.py

4. Run the iCub robot pronation-supination simulation:

python icub_new.py

5. Perform feature extraction:

python feature_engineering.py

6. Visualize features:

python features_visualization.py

7. Run MANOVA validation:

python manova_results.py

8. Train and validate CNN classifiers:

• For dataset with DBS intervention:

python cnn_with_dbs.py

• For dataset without DBS intervention:

python cnn_no_dbs.py

---

Repository Structure

.
├── MarmosetBG.py                  # Computational BG-T-C model simulation
├── icub_new.py                     # iCub robot pronation-supination simulation
├── feature_engineering.py          # Feature extraction scripts
├── features_visualization.py       # Feature visualization scripts
├── manova_results.py               # MANOVA statistical validation
├── cnn_with_dbs.py                 # CNN model training with DBS
├── cnn_no_dbs.py                   # CNN model training without DBS
├── requirements.txt                # Python dependencies

---

Acknowledgements

This work builds upon the foundation of Timothé Petitjean’s "Closed-Loop DBS for Parkinson's disease in the iCub" project. Extended contributions in feature engineering, statistical validation (MANOVA), and CNN-based severity classification have been added.

---