ECG Tensors and Visualization

Author: Vineet Tiruvadi 2019

A fun library I made during my IM rotation at Grady/Emory. It's meant to help visualize the learning process with ECGs, particularly with an emphasis on seeing the traditional ECG timetraces in a more modern, dynamical systems perspective. Maybe this will rely on other control theory-related repos I've got setup...

Dataset

We'll use the PhysioNet.org STAFFIII database.

Calculating augmented leads

The first step is to take the 9 channels from the STAFFIII database and compute our augmented leads. It's pretty straightforward to do, so we do it.

Phase portrait

ECGs are multiple measurements of a single underlying process: the heart beating. As it beats, it exhibits patterns that cardiologists have spent a lot of time mapping in detail. Those details, while important, can likely be simplified greatly by looking at the data in a different way. The engineering and physics fields give us a great way to view them: the phase space. In this space, we can see how variables relate to each other and how they change with respect to each other. With a very simple reframing of the data, we can see patterns emerge much more obviously.

An example of the phase space between channel V1 and V2 is displayed below

3D Phase portrait

Why not plot it in 3d? That's much cooler

Mechanical-Electrical (from scratch)

The next step is to build up a cardiomyocyte syncitium from scratch and try to get it to generate electrical signals in a forward model