Application of deep learning technology for temporal analysis of videofluoroscopic swallowing studies

Temporal parameters during swallowing are analyzed for objective and quantitative evaluation of videofluoroscopic swallowing studies (VFSS). Manual analysis by clinicians is time-consuming, complicated and prone to human error during interpretation; therefore, automated analysis using deep learning has been attempted. We aimed to develop a model for the automatic measurement of various temporal parameters of swallowing using deep learning. Overall, 547 VFSS video clips were included. Seven temporal parameters were manually measured by two physiatrists as ground-truth data: oral phase duration, pharyngeal delay time, pharyngeal response time, pharyngeal transit time, laryngeal vestibule closure reaction time, laryngeal vestibule closure duration, and upper esophageal sphincter opening duration. ResNet3D was selected as the base model for the deep learning of temporal parameters. The performances of ResNet3D variants were compared with those of the VGG and I3D models used previously. The average accuracy of the proposed ResNet3D variants was from 0.901 to 0.981. The F1 scores and average precision were 0.794 to 0.941 and 0.714 to 0.899, respectively. Compared to the VGG and I3D models, our model achieved the best results in terms of accuracy, F1 score, and average precision values. Through the clinical application of this automatic model, temporal analysis of VFSS will be easier and more accurate.