Deep Learning Model for Automating ECG Data Editing for Wearables

Burdulis, Vincenza; Sardak, Daniel; Nguyen, Jacob

Student Work

Deep Learning Model for Automating ECG Data Editing for Wearables

Public Deposited

Electrocardiograms (ECG) are frequently used to check the heart's health. Using electrodes, the electrical activity of the heart is able to be recorded. While this is usually performed by a professional, there are also wearable ECG devices. There’s different forms of the device, but each allows for an easy way to analyze the heart. A common problem with wearable electrocardiograms is that they aren't always accurate due to various reasons. To solve this issue a technician is able to manually edit the data to extract any noise or artifacts. Manually editing is a time consuming technique due to the large volume of data. Automating this process would significantly improve the efficiency and accuracy. This project develops a deep learning model to automate the data editing process for wearables. Our project, had tested out multiple machine learning models to see which had the highest performance. Through testing, we found that the Bidirectional Recurrent Neural Networks (Bidirectional RNN) and the Long Short Term Memory (LSTM) models had worked best with our data as they were able to detect over half the anomalies.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

Subject