A lightweight structure detector on cardiac ultrasound images of multiple views with tailored NMS algorithm

American Society of Echocardiography (ASE) has established guidelines for performing Transthoracic Echocardiographic exams in adults. 2D imaging protocols [1] are defined for multiple views that consist of several key structures. In practice, it can be difficult to determine whether qualified views have been achieved. In many cases, cardiac ultrasound images suffer from missing structures and incomplete chambers. This can lead to diagnostic errors and adversely alter patient treatment. Thus, we proposed a method that combines automatic cardiac view classification and detection of key structures with a tailored non-maximum suppression (NMS) algorithm. The multi-task method on different views is merged into a single light-weight model that has the potential to be deployed for real-time applications.