Physiological Signal Classification

Project title: Physiological Signal Classification

Leader: David Romero-Bascones (@drombas)

Collaborators: Stefano Moia

Topic: Machine learning, Physiology, Time-series analysis

Project description: Physiopy is is a python3 suite to format and analyse physiological recordings. One of the current development goals is to implementing an automatic signal classificator that, given a signal as input, is be able to determine the type of the signal.In this project we provide time-series data of 4 kinds of physiological signals (cardiac, respiratory chest, O2 and CO2) and the goal will be to collaborate to find robust features that allow discerning between them.

Data to use: Data to be used:

• 4 type of signals (cardiac, respiratory chest, O2 and CO2)- 240 time-series (60x4) recordings of 500 seconds long
• 2 files: time_series.csv (time-series data) and annotations.csv (time-series labels)

Link: https://www.dropbox.com/sh/3y5lhpn09qiz4my/AABKmpFuaGP_aHqxsAJ6LVwza?dl=0

Link to project repository: https://github.com/drombas/BHD-physiological-classification

Goals for Brainhack Donostia 2021: Easy:

• Develop a classifier between cardiac and respiratory signals (O2,CO2,chest) Doable but harder:
• Develop a classifier between all four signals Optimistic:
• Integrate the classifier into the development branch

First tasks: The project has 3 phases:

1. Exploration: download the data and visualize several signals to get an idea of how they look like.
2. Feature engineering: try to find features that discern the signals (using any programming language). You can create features:

• By hand: coming up with features based on intuition (we already have some ideas)
• Automatically with approaches like HCTSA (https://github.com/benfulcher/hctsa)

3. Implementation: build the final classification pipeline in Python with the best features.

Communication channels: https://mattermost.brainhack.org/brainhack/channels/physiopy

Video channel: Zoom

Number of collaborators: More

Credit to collaborators: Physiopy adopts the all-contributors system to recognise contributions. Contributors will be recognised as such in the relevant library README and as authors during outreach (conference posters, talks, …).

Type of project: Coding methods, Method development

Development status: Concept but no content

Programming languages: Julia, Matlab, Python, R

Necessary Programming skills level for the project: Familiar

Necessary git skills level for the project: None

Modality: ECG, physiology

Software suites: physiopy

Email: davidrb093g@gmail.com

What will participants learn: Basic time-series processing (normalization, filtering, maxima location, …). Time-series classification (feature engineering, validation, …).