ECG Analysis Application

We created an application for the automatic diagnosis of 12-lead ECG images.

## Background There are over 100 different cardiac conditions which are mostly diagnosed by interpreting 12-lead electrocardiogram (ECG) tests. We created a mobile application that can interpret the ECG test automatically, without any human intervention. The purpose of the application is not to replace the expert cardiologist's interpretation, but to serve as an assistant to medical personnel in providing the best immediate response and save lives by diagnosing conditions that require immediate action. While an expert cardiologist will know how to interpret the test, the technician that completes the exam or the nurse/doctor/physician that provides immediate care to the patient might not know how to interpret the ECG correctly. It can take a couple of hours until an expert cardiologist will see the test, and these hours can be critical in certain conditions.

Application for the automatic diagnosis of 12-lead ECG images.

*** ## Application Demonstration The application allows taking a video of an ECG test. The video is sent to our servers, where it is processed and the relevant frames are passed through a neural network that interprets the ECG. In less than 10 seconds, a result is sent back from the server to the application with the interpretation. The user can go back to past results and see the diagnosis of the machine for each ECG, along with an explainability image that shows the locations on the ECG test which the neural network gave more weight to. Screenshots of the application:

Demonstration of the application.

## Diagnosed Diseases We diagnose a total of 23 different cardiac conditions, with accuracy scores of 85% or higher. ## Clinical Trial The application is currently being used by several doctors in different hospitals around the world. We are gathering new data that will help us to automatically diagnose new cardiac conditions and also improve the accuracy of our results. For the clinical trial, we have received approval from the Helsinki committee. ## Discussion We created a neural network that can diagnose different cardiac conditions, based on 12-lead ECG images. Our model currently needs a couple of hundred images of a cardiac condition to successfully diagnose a new cardiac condition. We are researching ways in which we can reduce this number while still getting high accuracy results. The clinical trial will help us to gather new data, and also test the performance of our model on unseen data.