Fig. 2.

Overview of the model architecture of the Transfer Esophagectomy Network (TEsoNet): The individual video frames from laparoscopic Sleeve Gastrectomy and the laparoscopic part of Ivor-Lewis Esophagectomy are loaded in to a convolutional neural network (CNN). A following fully connected layer transforms the output for loading into a long short term memory (LSTM) model. Additional transformation through a second fully connected layer reduces the output to 10 classes before undergoing a rectified linear activation function (ReLU). The output is a probability for each individual class (phase)