nirvanesque/OMNI

Provides Open Source biosignal Deep Learning models for robust monitoring of neonate breathing and cardiac health along with a well documented wearable edge implementation guide.

OMNI (Open Source Monitoring of Neonates and Infants)

Software Requirements

Run sh requirements.sh in a virtual environment in order to download the required libraries.

Hardware Requirements (Edge implementation)

Components needed:

1. OpenBCI Ganglion kit

   Open source biosignal acquisition hardware for research grade biosignal acquisition

2. Raspberry Pi 4

3. Rasberry Pi 4 cooling case

4. Shirt using conductive textile electrodes [1$ for 1 electrode]

Wearable ECG electrodes --> OpenBCI Ganglion ---(Bluetooth)---> Raspberry Pi 4

System Configuration

• Ubuntu 16.04
• Nvidia 1080Ti - (Required for training the model)

Train a model to extract R peaks and Heart Rate from ECG waveform.

• Download ECG MITDB monitoring data from https://storage.googleapis.com/mitdb-1.0.0.physionet.org/mit-bih-arrhythmia-database-1.0.0.zip and unzip it.
• To train: python train.py --preprocess_data --data_path "PATH TO DATA"

Train a model to extract Breathing Rate from ECG waveform.

• Download ECG from the BIDMC database which is derived from MIMIC-II from https://physionet.org/static/published-projects/bidmc/bidmc-ppg-and-respiration-dataset-1.0.0.zip and unzip it.
• To train: python train.py --preprocess_data --data_path "PATH TO DATA"

Model Inference

• Download ECG from the preterm infant database from https://physionet.org/static/published-projects/picsdb/preterm-infant-cardio-respiratory-signals-database-1.0.0.zip and unzip it.
• Download the Heart Rate computation model from here: https://drive.google.com/open?id=1yI7G4nofjuzFWkD1CfsOtLZxaukTu0di
• Download the Breathing Rate computation model from here: https://drive.google.com/open?id=1ycV74LfGmgcGmLlrPn2VileeFNsGrRZT
• To run inference and view GUI type: python run_model.py --path_dir "PATH TO DATA" --saved_hr_model_path "PATH TO HR MODEL" --saved_br_model_path "PATH TO BR MODEL" --patient_no 3 --viewer 1

OMNI OpenBCI Pi Inference

Edge inference of ECG R-peak detection and Respiration extraction using Raspberry Pi 4 using ECG (OpenBCI Ganglion).

Hardware Design

Wearable ECG electrodes --> OpenBCI Ganglion ---(Bluetooth)---> Raspberry Pi 4

Software Design

OpenBCI client ----(LSL)---> Python -> PyTorch inference --> Breathing Rate, Heart Rate

Installation instruction

Install PyTorch on Raspberry Pi 4:

1. Install PyTorch dependicies sudo apt install libopenblas-dev libblas-dev m4 cmake cython python3-yaml libatlas-base-dev
2. Increase swap file memory to 1600, Edit variable CONF_SWAPSIZE in /etc/dphys-swapfile
3. Reset environmental variables like ONNX_ML Instructions
4. Download PyTorch package compiled for Armv7 (torch-1.1.0-cp37-cp37m-linux_armv7l.whl)
5. Install using the command sudo pip3 install torch-1.1.0-cp37-cp37m-linux_armv7l.whl in the same directory

Refer here for troubleshooting

Install OpenBCI Ganglion client on Raspberry Pi 4:

1. Clone OpenBCI_Python repo git clone htps://github.com/OpenBCI/OpenBCI_Python.git
2. Install the following requisites python packages using pip3 install pylsl, python-osc, six, socketIO-client, websocket-client, Yapsy, xmldict, bluepy
3. Open folder OpenBCI_Python and run
   sudo python3 user.py --board ganglion -a streamer_lsl to open a lab streaming layer stream of sensor data from the ganglion

Instructions to run to perform real time breathing rate/ heart rate inference using OpenBCI data

1. Run the lsl streamer script to get data to the inference script sudo python3 user.py --board ganglion -a streamer_lsl
2. Run the visualization and edge inference code on the pi using python3 lsl_openbci.py

Sample Predictions

Above is an example prediction for noisy real time ECG data obtained using the edge inference model. The beat predictions are represented as blue markers on the ECG.