Echocardiograms are ultrasounds of the heart that are used primarily in cardiovascular disease diagnoses. Echos can be used to pinpoint the presence of disease that could not otherwise be physiologically detected. With new advancements in both computer vision and high performance computation, we can now apply machine learning algorithms to echo images in order to automatically detect areas of interest.

Our project built a system where image analysis of echos occurs in real time. Using the power of computational neural networks, we trained a model that detects ventricles and predicts ventricular volumes. Then, we created a stream of input data that emulates the function of an ultrasound probe. This input was then run through our generated model, and displayed in real-time to a user application. The end result was an image with an overlay marking of the detected ventricles, as well as predicted ventricular volume.

In future iterations of this project, more fine grained analysis for specific diseases or other metrics can be done. All in all, our project provides a solid framework for further exploration of machine learning with echos.

The group would like to thank their faculty sponsor David Ferry for their support of this project.