2021 Volume Annual59 Issue Abstract Pages 432
Background: The objective of this study is to assess the feasibility to apply deep learning-based spatiotemporal video analysis using a three-dimensional convolutional neural network (3D-CNN) to automatic surgical skill assessment (ASSA).
Methods: Laparoscopic colorectal surgical videos with reliable skill assessment scores were collected from the Japan Society for Endoscopic Surgery, and among them, 74 laparoscopic colorectal surgical videos were utilized for this study. A total of 1480 video clips were extracted, and divided 4:1 for the training and the test set, respectively.
Results: The proposed 3D-CNN model could automatically classify video clips into 3 different score groups (i.e., <Average-2SD, Average±SD, and >Average+2SD) with 75.0% accuracy and screen for <Average-2SD score group with 94.1% sensitivity and 96.5% specificity.
Conclusions: 3D-CNN, which can analyze videos instead of static images had the potential to be applied to ASSA.
