Introduction

Currently, Point-of-Care (POC) ECG monitoring works either as plot devices and/or alarms for abnormal cardiac rhythms using predefined normal trigger ranges. On the other hand, full 12-derivation ECG machines are complex to use as simple monitors and are used with strict techniques for formal diagnostics of hearth electric conduction pathologies, and the automatic diagnostics are derived from a full analysis of the 12-dimension data after it is fully collected. In CinC/Physionet Challenge 2015, it has been reported that up to 86% resulting of the alarms are false and this can lead to decreased staff attention and increase in patients delirium (Chambrin, 2001; Parthasarathy & Tobin, 2004).

Research question and aims

This research aims to identify, on streaming data, abnormal hearth electric patterns, specifically those who are life-threatening, in order to be a reliable signal for Intensive Care Units to respond quickly to those situations. It also may be able to continuously analyze new data and correct itself shutting off false alarms. Primarily an experiment will be conducted using 2 main algorithms that use Matrix Profile in detecting context changes: SDTD and FLOSS. One uses whole data training and testing, and the other uses a streaming approach that is our main interest. The goal will be detecting the transition from normal to flutter/FA to normal condition with special attention to not rely on rhythm changes. Being this successful, a more generalistic approach will be attempted: to detect changes from normal to abnormal to normal conditions, with special attention to handle with disconnected leads or patient movements. Finally, this research can prove to be a good addition to the Matrix Profile method, using fading factors in order to reduce memory and space consumption, lowering the processor power needed, allowing this algorithm to be used in almost any device.

About the ongoing project

The document submitted for approval is here.

The full code is open-sourced and available here

To follow the thesis timeline you can access the full Gantt chart at Zenhub. Click here (you need a github account, but that’s it).

PDF, EPUB and WORD versions will be available at the end of this work.