Fig. 2From: Real-time estimation of the remaining surgery duration for cataract surgery using deep convolutional neural networks and long short-term memoryStructure of Proposed Method. a The proposed regression model which consists of CNNs and LSTM. CNNs are used to extract the features for each frame. LSTM will analyze the feature in time order and output the process (%) prediction for each elapsed time. b The figure on the left is the real-time prediction of a sample from surgeon 1’s test set. The horizontal axis is the true elapsed time (s), and the column axis is the progress (%) prediction of where the current time is during the whole video. The green line is the true label and the red line is the prediction. The figure on the right is the overlap map for observing the prediction error for the whole test set. The horizontal axis is the true elapsed time (s), and the column axis is the prediction error (s). The overlap map is drawn by overlapping the prediction error curve of all the test set samples of surgeon 1Back to article page