Doctoral Symposium - EPIA 2022 - Francisco Bischoff

Detecting life-threatening patterns in Point-of-care ECG using efficient memory and processor power



Author: Francisco Bischoff


Center of Health Technology and Services Research

Faculty of Medicine of Porto University

HEADS - PhD. Programme in Health Data Science

2022-08-30

The problem

ECG

  • Simple and straightforward concept for our time.

  • Still requires high-quality devices.

  • Automatic interpretations are not good enough.

  • Still requires lots of (human) training for correct interpretation.

  • Research field still being expanded, but usually for improve accuracy, not for expanding usage scenarios.

Monitoring

  • The rule is one complex monitor per patient and medical evaluation.
  • Outpatient monitoring usual is based on Holter devices.
  • Analysis is made afterwards, offline.
  • The equipment is not cheap, and the user is not the “general population”.

Monitoring

  • In our days we are collecting large amounts of data from ourselves.
  • There are wearable devices powerful enough to read and analyse ECG in real-time.
  • The general population could benefit from specific life-threatening warnings.

Objectives and research question

Objectives and research question

  • Objective

    Detect life-threatening patterns in Point-of-care ECG
    using efficient memory and processor power.

  • Main question

    Can we accomplish this objective while maintaining robustness?

Planned approach

Overview

Overview of the final implementation.

Matrix Profile

State of art time-series analysis [3]

The concept of Matrix Profile

FLOSS

Regime change detection [1]

Explaining the arc counts

FLOSS

FLOSS

Example of regime change

Classification

Contrast Profile [2]

Finding class signatures with contrast profile

Classification

Snippets candidates

Snippet candidates extracted using the contrast profile

Questions



This presentation:
https://rpubs.com/franzbischoff/epia2022
Project website:
https://franzbischoff.github.io/false.alarm

References

Detecting life-threatening patterns in Point-of-care ECG using efficient memory and processor power

1.
Gharghabi, S. et al.: Domain agnostic online semantic segmentation for multi-dimensional time series. Data Mining and Knowledge Discovery. 33, 1, 96–130 (2018). https://doi.org/10.1007/s10618-018-0589-3.
2.
Mercer, R. et al.: Matrix profile XXIII: Contrast profile: A novel time series primitive that allows real world classification. In: 2021 IEEE international conference on data mining (ICDM). pp. 1240–1245 (2021). https://doi.org/10.1109/ICDM51629.2021.00151.
3.
Yeh, C.-C.M. et al.: Matrix profile i: All pairs similarity joins for time series: A unifying view that includes motifs, discords and shapelets. In: 2016 IEEE 16th international conference on data mining (ICDM). pp. 1317–1322 IEEE (2016). https://doi.org/10.1109/ICDM.2016.0179.