RSDNet: Learning to Predict Remaining Surgery Duration from Laparoscopic Videos Without Manual Annotations

Objective: Accurate surgery duration estimation is necessary for optimal OR planning, which plays an important role for patient comfort and safety as well as resource optimization. It is however challenging to preoperatively predict surgery duration since it varies significantly depending on the patient condition, surgeon skills, and intraoperative situation. We present an approach for intraoperative estimation of remaining surgery duration, which is well suited for deployment in the OR. Methods: We propose a deep learning pipeline, named RSDNet, which automatically estimates the remaining surgery duration intraoperatively by using only visual information from laparoscopic videos. An interesting feature of RSDNet is that it does not depend on any manual annotation during training. Results: The experimental results show that the proposed network significantly outperforms the method that is frequently used in surgical facilities for estimating surgery duration. Further, the generalizability of the approach is demonstrated by testing the pipeline on two large datasets containing different types of surgeries, 120 cholecystectomy and 170 gastric bypass videos. Conclusion: Creation of manual annotations requires expert knowledge and is a time-consuming process, especially considering the numerous types of surgeries performed in a hospital and the large number of laparoscopic videos available. Since the proposed pipeline is not reliant on manual annotation, it is easily scalable to many types of surgeries. Significance: An improved OR management system could be developed with RSDNet as a result of its superior performance and ability to be efficiently scaled up to many kinds of surgeries.