Fig. 2.

a Implemented U‑Net inspired neural network architecture. For each layer of the network, the height of a block represents width and height in pixels of the input and output images, and the size of data in memory in each of the model’s layers. The width of each block represents the depth, or the number of parallel filters through which data pass in each layer of the network. Act type of activation function; ReLu rectified linear unit, or rectifier, an activation that outputs zero for inputs less than zero and outputs that equal their input for inputs greater than or equal to zero. b Left ventricular ejection fraction (LVEF) determined from the contours predicted by the U‑Net model compared to the LVEF derived from manual annotations