Apnea/hypopnea detectors for polysomnographic data, specifically for the PhysioNet 2018 dataset.
One of the two detectors works on the base of classical signal-processing and rule-based decisions. The other detector makes use of modern AI-based methods.
Important: Before starting anything, make sure you install all necessary dependencies by preparing the Conda environment, see steps down below.
In both cases (rule-based & AI-based detector), a good start is to take a look at the provided notebooks under the equally named sub-folder. They demonstrate how to use the data processing infrastructure, how to plot nice images and how to eventually run the apnea/hypopnea detectors.
However, it is recommended to place the PhysioNet datasets into the sub-folder data. The
therein included README.md provides more information. The aforementioned
notebooks will make use of the files stored in that folder.
For those who are interested in training own AI models on the PhysioNet dataset: You
should take a look at the files within sub-folder ai_based, most of all at the contained
README.md file.
The AI training framework contained in sub-folder ai_based was largely built on top of
TheFloe1995's GitHub repo (MIT license), which
provides a nice way to train and manage AI models as so-called experiments.
- conda env create --file environment.yml
- conda activate apnea-detection
- conda install -y pytorch cudatoolkit=11.1 -c pytorch -c nvidia