Fig. 2: PINN training results of 3D vocal fold dynamics in the canine larynx.
From: Predicting 3D soft tissue dynamics from 2D imaging using physics informed neural networks
a The history of data (left) loss and equation (right) loss during the training. b The prediction error of 3D vocal fold shapes over one cycle. The dashed line and the shaded area indicate mean and standard deviation (SD) over a cycle, respectively. c The mean and SD of the prediction error of 3D vocal fold shapes versus the number of eigenmodes adopted. d Comparison of 3D vocal fold shapes and vertical velocity component contour between the PINN prediction and ground truth at 5 representative time instants in a cycle (shown in Fig. 3a). e Comparison of the vertical profile of the vocal fold at the mid-coronal plane between the PINN prediction and ground truth at the corresponding time instants.
