Deep Learning Pipeline for Echocardiogram Noise Reduction

The echocardiogram is an imaging modality based on ultrasound imaging of the heart and neighboring regions for assessing heart functions. It is employed to understand how the heart chambers and valves pump the blood and can be used to detect heart failures, heart valve problems, and other cardiac abnormalities. The ability to detect such abnormalities from echocardiogram images is based both on the skill of cardiologists and on the quality of the images. Due to reasons such as body fat, positioning of the ultrasound probe, chest hair, presence of tissues and bones, etc... The echocardiogram images often have a lot of noises like speckle noise and shadowing which affects the final diagnosis and at times results in missed diagnosis. In this work, we have developed a latent space representation-based technique to remove noise from echocardiogram images specifically for the particular challenge of identifying regional wall motion abnormality. The analysis of our algorithm on a dataset of 53 patients, demonstrated the performance of our algorithm in reducing noise in such images.