NDDNet

A Deep Learning Model for Predicting Neurodegenerative Diseases from Gait Pattern

Implementation of the NDDNet Architecture

The TensorFlow implementation of the architecture can be found here.

Figure: The proposed network architecture. The network is composed of two parallel parts for processing the vGRF signals and the gait-cycle parameters. (a) The entire network architecture. (b) The ConvMixer block architecture.

Dependencies

  numpy
  pandas
  scikit-learn
  tensorflow
  wfdb
  tsfel
  rich

Dataset

The dataset used in this work can be found on PhysioNet. Its also available in the data/gaitndd directory.

How to Use

1. Install the dependencies
2. Run src/Inference.py for doing the inference on the dataset
3. Run src/train.py to train the model

Pre-trained Weights

Pretrained weights are available in the weights directory. The weights were generated by a (Leave-One-Out-Cross-Validation) LOOCV method.

Cite This Work

Faisal, M.A.A., Chowdhury, M.E.H., Mahbub, Z.B. et al. 
NDDNet: a deep learning model for predicting neurodegenerative diseases from gait pattern. 
Appl Intell (2023). https://doi.org/10.1007/s10489-023-04557-w

Qatar University Machine Learning Group