Fig. 1
From: Automatic cardiac evaluations using a deep video object segmentation network
The workflow of the algorithm for LV series. After pre-processing, the first frame and its supervision enter into the reference stream, and the following frames enter into the main stream one after the other after pre-processing. The inputs are passed through the Siamese encoder, FPN, and three final subnets (classification, regression, and segmentation subnets), which are the main components of EchoRCNN. Then, the main parameters of LV functionality (length, area, and volume) are extracted after post-processing on the predicted masks. By having these parameters for each frame, end-diastolic and end-systolic frames are detected, and finally, the EF parameter can be calculated.