Left Ventricular Assist Device Thrombus Detection Using Wavelets and Image Classification

Abstract

Cardiac related diseases are a common health risk for adults. Consequently, therapies such as heart transplants and medication exist to treat these ailments. Heart transplants remain the gold standard for treating severe heart failure, however, left ventricular assistive devices, a cardiac blood pump, are gaining popularity and not just as a bridge for long term care. Unfortunately, with the benefits of these devices come risks of clot formation. These occlusions can cause strokes, further cardiac damage, or even death. Therefore, these occlusions must be detected at the onset. This work presents a method to non-invasively monitor the condition of a Thoratec HeartMate II ventricular device. The application of a neural network and a classification tree are designed to detect the presence of an aortic graft occlusion that has been seeded into an in vitro cardiac simulator. Using acoustic digital heart sounds, the classification tree showed the most favorable results, outperforming the existing support vector machine method by roughly 20%.