29.09.2016, Pawel alxd Chojnacki and Ryszard Cetnarski
NeuroOn Hypnogram Analysis
Following my principles of Open Access and Open Notebook Science I'd like to present you my comparative analysis of NeuroOn and a professional polisomnograph recording from two nights. To learn more about the experiment itself, please head back to my previous blog post.
Raw signals are available for download here, with md5sum
NeuroOn signal was obtained by using proprietary Intelclinic's scripts I am not allowed to share. They produced four CSV files, containing respectively:
- EEG signal (125 Hz)
- accelerometer signal
- LED activity
- Staging (hypnogram)
AURA PSG signal was exported to an EDF format using Grass Technologies PSG TWin 4.5.4 and 4.5.2 to minimize the risk of software-derived artifacts. The channels signals include:
- SNORE (first night only)
- FLOW (first night only)
- CHEST (first night only)
- ABDOMEN (first night only)
- ECG (first night only)
- SaO2 (blood oxygen saturation, first night only)
- HR (heart rate, first night only)
- LEG1 (first night only)
- LEG2 (first night only)
- Pos (first night only)
All EEG channels were sampled with 200 Hz frequency. PSG signals from both nights were scored by a professional and exported to XLS and CSV files.
This is a peer-review version of the analysis (currently) with no clear conclusions or discussion section. Please refrain yourself from making hasty assumptions related to NeuroOn product before it is read and commented by third party researchers. We would like to discuss the science behind the research, not the market viability of the device examined.
The goal of the experiment is to detect time shift between NeuroOn and Aura PSG, each of which use different clock and compare synchronized hypnograms. In order to test NeuroOn's accuracy it is assumed that PSG signal is a source of truth.
Time_synchronization is the first notebook analyzing time shifts between NeuroOn and PSG signals and hypnograms.
Hipnogram_comparison uses Cohen's cappa to measure NeuroOn's accuracy compared to the PSG.
Spectral_analysis compares PSG's and NeuroOn's ability to use delta power to determine sleep stages.
The analysis doesn't include research in NeuroOn's use of accelerometer, focusing on EEG stage detection only. Continuing in this direction (especially given additional signal gathered from the bicep) may grant additional insight into NeuroOn's functioning.
It is strongly advised to use
virtualenv for freezing specific version of Python libraries. Please use
pip 8.12 . Instructions:
virtualenv(what is it?) in core project folder and activate it.
pip install -r requirements.txtto install all dependencies. It may be required to update your package installer with
pip install --upgrade pip.
- Download hdf files and unzip them to populate the
- Download raw signal files and unzip them to populate the
jupyter notebookto start the notebook.
All feedback welcome
Feel free to contact me, fork and share this repository!
All the code and signal samples are provided under MIT license (see LICENSE.md). Other resources, such as blogposts and scientific posters may have different licenses, such as viral CC-BY-SA.