Skip to content

Latest commit

 

History

History
23 lines (20 loc) · 3.56 KB

README.md

File metadata and controls

23 lines (20 loc) · 3.56 KB
Raw

Sleep-disorder-Classification

A hierarchical approach for the diagnosis of sleep disorders and cyclic alternating pattern (CAP) sleep phases using 1-dimensional convolutional recurrent neural network (CRNN). The hierarcy has three stages, first stage classifies input signal into healthy and unhealthy. The next stage identifies sleep disorders from unhealthy input signals namely insomnia, nfle, narcolepsy, rbd and plm. The third stage identifies CAP phase of input signal. Proposed models uses single-channel standardized electroencephalogram (EEG) recordings provided by the CAP sleep database. No manual feature extraction or pre/post processing is required making the approach completely autonomous. The healthy-unhealthy classification and disease classification have been studied using dataset of both phases (A & B), using dataset of only phase A, using dataset of only phase B seperately.

Instructions

Dataset preparation

  • Follow instructions of our repository CAP-Phase-Detection to create balanced CAP dataset for healthy as well as disordered subjects. You can download this data directly from here.
  • Run the colab file Dataset_preparation_disease_classification. Give proper path of previously downloaded dataset and for folder in which dataset is to be strored.
  • You can directly download dataset for healthy-unhealthy classification from here and dataset of disease classification from here

Model training

  • Run the colab file Healthy_Unhealthy_Classification.ipynb for training model for first stage of hierarcy using dataset of both phases, phase A and phase B and observe result.
  • You can observe results of our proposed models by downloadig our trained models, history and test datasets from here.
  • Run the colab file Disease_Classification.ipynb for training model for second stage of hierarcy using dataset of both phases, phase A and phase B and observe result.
  • You can observe results of our proposed models by downloadig our trained models and datasets from here.
  • You can refer the colab file Hyperparameter_optimization.ipynb for optimizing the hyperparameters of models.

GUI

  • Download the folder GUI to your local pc.
  • Run the run.py file using python interpreter.
  • Open the generated link in any browser.
  • Give the test samples provided in the folder test or any C4A1 channel signal of 2s duration of 512Hz in csv format to see the result.

The sample output of GUI screen can be seen below