Parkinson's Disease and Sleep

This study aims to study the correlation between chronotype, sleep quality and Parkinson's disease.

Sebastiano Vacca, Taha Binhuraib

Dependencies

• OS: Linux or macOS
• Language: Python
• Libraries: scikit-learn, numpy, pandas

Installation

Git Clone

Git clone the repository by running the following commands:

# change directory to Documents
cd Documents
git clone https://github.com/TahaBinhuraib/pd_sleep.git
cd pd_sleep

Anaconda

You need to install Anaconda, then run below:

conda create --name pd python=3.8
conda activate pd
pip install -r requirements.txt

Usage

Jupyter Notebook

• Random Forest
• Data Analysis

Python 🐍

You can run our python deep and machine learning model by running the following:

cd model
# random_forest 
python random_forest.py
# Autoencoder
python autoencoders.py

Evaluation

Voxel-Based-Morphometry(VBM)

VBM has been conducted on MATLAB, with SPM12 & CAT12 toolboxes. The results were visualized with xjview. We used both Grey Matter and White Matter segmented images, after smoothing with a 8x8 Kernel, as inputs for our two-way t-test between Parkinson Patients and Healthy Controls. A p.value of < 0.001 and a Family Wise Error of 0.05 have been used. Regarding the GM analysis, the most significant differences have been found in the putamen, both left and right ones, with the HC having more GM. As far as the WM analysis, the most significant differences have been found in the sub-gyral parietal lobe, with the HC having slightly less WM.

Source-Based-Morphometry(SBM)

SBM consists in applying PCA/ICA (through the infomax algorithm) to the smoothed images to have a better visualization and analysis of GM and WM densities. SBM has been conducted on MATLAB, with the FUSION toolbox. The inputs are the same of the VBM. Firstly, we ran the analysis with only one feature, the GM, and then with both GM and WM. Only the components with a p.value of <0.001 have been reported here. The regions have been identified based on the Talairach Coordinates, where the images have been converted to Z-scores with a threshold of 2.

Source-Based-Morphometry(SBM) - Part 1

SBM consists in applying PCA/ICA (through the infomax algorithm) to the smoothed images to have a better visualization and analysis of GM and WM densities. SBM has been conducted on MATLAB, with the FUSION toolbox. The inputs are the same of the VBM. Firstly, we ran the analysis with only one feature, the GM, and then with both GM and WM. Best component for each feature combination is deter- mined using two sample t-test on the mixing coefficients. Only the components with a p.value of <0.05 have been reported here. The regions have been identified based on the Talairach Coordinates, where the voxels above threshold |Z| > 1.5 were converted.

GM ONLY

7th Independent Component(IC): The Lentiform Nucleus is the area where the HC show the highest GM density compared to PD, while the Precentral Gyrus the lowest.

GM&WM

1st IC & Feature 1: The Lentiform Nucleus is the area where the HC show the highest GM density compared to PD, while the Middle Frontal Gyrus the lowest.

2nd IC & Feature 1: Same as the 1st as far as the two regions with the most differences go.

1st IC & Feature 2: The Lingual Gyrus is the area where the HC show the lowest WM density compared to PD, while the Middle Frontal Gyrus the highest.

2nd IC & Feature 2: Areas 20 and 21 is where the HC show the lowest WM density compared to PD, while the Fusiform Gyrus the highest.

Source-Based-Morphometry(SBM) - Part 2

The analysis consisted in comparing the Grey and White Matter of 11 PD patients (median age 39 years) vs 11 Healthy controls with an Early Chronotype (EC - median age 31 years) and other 11 with a Late Chronotype (LC - median age 31 years). Only the components with a p.value of <0.05 have been reported here. The regions have been identified based on the Talairach Coordinates, where the voxels above threshold |Z| > 1.5 were converted. The regions with the maximum t-value have been reported.

GM ONLY

PD vs EC

7th Independent Component(IC): The Middle and Inferior Temporal Gyrus are the areas where the EC show the highest GM density compared to PD, while the Uncus the lowest.

2nd Independent Component(IC): The Superior and Middle Frontal Gyri are the areas where the EC show the highest GM density compared to PD, while the Precentral Gyrus the lowest.

PD vs LC

7th Independent Component(IC): The Middle and Inferior Temporal Gyrus are the areas where the LC show the highest GM density compared to PD, while the Uncus the lowest.

5th Independent Component(IC): The Angular Gyrus and Parahippocampal Gyrus are the areas where the LC show the highest GM density compared to PD, while the Middle and Inferior Frontal Gyri the lowest.

GM & WM

PD vs EC

8th IC & GM: The Lentiform Nucleus and Brodmann area 13 are the areas where the LC show the highest GM density compared to PD, while the Cerebellar Tonsil and the Superior Frontal Gyrus the lowest.

8th IC & WM: The Medial Frontal Gyrus and the Sub-Gyral Brodmann areas 6,32, 37 and 40 are the areas where the LC show the highest GM density compared to PD, while the Middle Temporal Gyrus and the Sub-Gyral area 8 the lowest.

PD vs LC

4th IC & GM: The Sub-Gyral Brodmann areas 6, 30 and 40 are the areas where the LC show the highest GM density compared to PD, while the Middle Temporal Gyrus and the Supramarginal Gyrus the lowest.

4th IC & WM: The Sub-Gyral Brodmann areas 6, 7, 8 and 21 and the Middle Occipital Gyrus are the areas where the LC show the highest GM density compared to PD, while the Inferior and Superior Parietal Lobule the lowest.

7th IC & GM: The Thalamus and Brodmann areas 6, 20, 21 and 40 are the areas where the LC show the highest GM density compared to PD, while the Lingual Gyrus and the Inferior Occipital Gyrus the lowest.

7th IC & WM: The Medial Frontal Gyrus and the Inferior Parietal Lobule are the areas where the LC show the highest GM density compared to PD, while the Lingual Gyrus and the Middle Frontal Gyrus the lowest.