SafwenNaimi / HCT-Hybrid-Convnet-Transformer-for-Parkinson-s-disease-detection-and-severity-prediction-from-gait

HCT: Hybrid Convnet-Transformer for Parkinson’s disease detection and severity prediction from gait

Prerequisites

Python 3.8 or later with all requirements.txt dependencies installed. To install run:

python -m pip install -U pip

pip install -r requirements.txt

Dataset

The dataset used in the paper is from Physionet. It can be downloaded from (1). A sample from each group of researchers contributed in the dataset is available in the subfolder data. Ga, Ju or Si – indicate the study from which the data originated:

• Ga - Galit Yogev et al (2) (dual tasking in PD; Eur J Neuro, 2005)
• Ju – Hausdorff et al (3) (RAS in PD; Eur J Neuro, 2007)
• Si - Silvi Frenkel-Toledo et al (4) (Treadmill walking in PD; Mov Disorders, 2005)
• Co or Pt: Control subject or a PD Patient

We also included demographics.xls that contains all the information about each example in the data. Each line in demographics.xls contains 19 columns:

• Column 1: Time (in seconds)
• Columns 2-9: Vertical ground reaction force (VGRF, in Newton) on each of 8 sensors located under the left foot
• Columns 10-17: VGRF on each of the 8 sensors located under the right foot
• Column 18: Total force under the left foot
• Column 19: Total force under the right foot.

Getting Started

To run experiments for our Two-class ConvNet-Transformer model, the entry point is Two-Class_model.py file.

The algorithm will generate the following output files:

├── output(dir)
    ├── train_classifier_month_day(dir)   
        ├── hour_minutes(dir)
            ├──  model.json : JSON file of the model.               
            ├──  res_pat.csv: results of accuracy, sensitivity and specificity by patients.
            ├──  res_seg.csv: results of accuracy, sensitivity and specificity by segments.	                
            ├──  training_i.csv: training/validation loss and accuracy for the i_th folder (i = [1..10]).   
            ├──  weights_i.hdf5 : weights of the model for the i_th folder (i = [1..10]).   

To run experiments for our Multi-class ConvNet-Transformer model, the entry point is Multi-Class_model.py file.

The algorithm will generate the following output files:

├── output (dir)
    ├── train_severity_month_day(dir)   
        ├── hour_minutes(dir)
            ├──  confusion_matrix.csv: Confusion matrix for severity prediction.
            ├──  gt.csv: ground truth level for each patient.
            ├──  pred.csv: prediction level for each patient.
            ├──  model.json : JSON file of the model.               
            ├──  res_pat.csv: results of accuracy by patients.
            ├──  res_seg.csv: results of accuracy by segments.	                
            ├──  training_i.csv: training/validation loss and accuracy for the i_th folder (i = [1..10]).   
            ├──  weights_i.hdf5 : weights of the model for the i_th folder (i = [1..10]).   

References:

(1): https://physionet.org/content/gaitpdb/1.0.0/

(2): G. Yogev, N. Giladi, C. Peretz, S. Springer, E. S. Simon, and J. M. Hausdorff. “Dual tasking, gait rhythmicity, and Parkinson’s disease: Which aspects of gait are attention demanding?” In: European Journal of Neuroscience 22 (2005).

(3): J. M. Hausdorff, J. Lowenthal, T. Herman, L. Gruendlinger, C. Peretz, and N. Giladi. “Rhyth- mic auditory stimulation modulates gait variability in Parkinson’s disease”. In: European Journal of Neuroscience 26 (2007).

(4) S. Frenkel-Toledo, N. Giladi, C. Peretz, T. Herman, L. Gruendlinger, and J. M. Hausdorff. “Treadmill walking as an external pacemaker to improve gait rhythm and stability in Parkin- son’s disease”. In: Movement Disorders 20 (2005).