Week 6 - Alex

This week I began work on developing a neural net that can detect premature ventricular contractions from an ECG.

The clinical research project that I’m working on requires that we identify, from 24 hour ECG data, whether a patient with PVCs is indicated for treatment with Adenosine. Currently, there is no known way to distinguish between patients with different PVC mechanisms. Some groups posit that the variance of the coupling interval between the previous QRS complex and the PVC could indicate an underlying mechanism: 

• Patients with a low variance in their coupling interval are thought to have PVCs originating from modulated parasystole
• Patients with a high variance are thought to have ectopic PVCs, originating in the right ventricular outflow tract

The clinical trial hypothesizes that adenosine will be an effective treatment for patients in the latter group. This hypothesis is based on an assumption that RVOT PVCs are mechanistically similar to ventricular tachycardias, which is treated by Adenosine.

I have implemented a neural net that is able to identify PVCs in ECGs downloaded from the MIT-BIH ECG database. Since I am able to find the location of the PVCs, I can determine the coupling interval for all the PVCs in a 24 hour time sequence. I will pursue two approaches for clustering patient data:

1. Calculate the coupling interval variance for all PVCs in each patient
2. Perform a principal component analysis on PVCs from each patient to determine if there is a principal component that clusters patients into two different groups.

In addition to these approaches, I will soon begin work with a Dr. Peter Oken and Professor Olivier Elemento on a project for analyzing ECG data using machine learning approaches. My hope is that I can apply these methods to make insights into the ECG data.