Measurement of the pediatric right ventricle is an important clinical indicator for the health and development of pediatric patients. Clinicians leverage echocardiography to acquire safe, inexpensive, and dynamic images of the heart. Contouring, or segmentation, of the right ventricle in echocardiograms is integral to multiple measurements of cardiac health, such as fractional area change, strain, and volume. Automatic segmentation of the right ventricle has many benefits, like increasingly consistent and high resolution tracking of pediatric cardiac health, however developing these models comes with many challenges due to inconsistency in human annotation and variability in image quality. This work details deep learning methods and pitfalls for pediatric right ventricle segmentation in echocardiography, along with examples to illustrate these findings, along with a method for quantifying predicted segmentation uncertainty as well as methods for data and model evaluation using active feedback from cardiologists and sonographers. This work will enable application and extension of methods for this task and related, challenging video–based segmentation tasks.