A real-world study on establishing an artificial intelligence-based laparoscopic cholecystectomy final total operative duration prediction model to improve the operational efficiency of the operating room

Artificial intelligence (AI) has shown promise in various clinical applications. However, its application to improve the efficiency of laparoscopic cholecystectomy operating rooms remains underexplored. This was a two-phase, non-randomized, quasi-experimental study with a historical control group. In Phase I, a group of 145 patients who underwent laparoscopic cholecystectomy under the traditional mode was used to develop a Long Short-Term Memory Network (LSTM) model for predicting the min-max-normalized final total operative duration for each surgical case, with inverse transformation to minutes for clinical display. In Phase II, the AI system was deployed in 62 consecutive patients as the AI-Assisted intervention and temporal test cohort, and its data were not used for model training, hyperparameter tuning, or model selection. The model’s end-of-sequence predictive performance was assessed by the Root Mean Square Error (RMSE) to evaluate the error between inverse-transformed predicted and actual final total operative duration values in minutes. Multiple clinical outcomes were compared between the Traditional group ( n = 145) and the AI-Assisted group ( n = 62), including first-surgery delay rate, surgical turnover time, patient waiting time, patient satisfaction, and surgeon satisfaction. The LSTM model yielded an end-of-sequence RMSE of 0.593 min in the Phase I development-validation subset and 0.637 min in the Phase II temporal test cohort. Compared with the Traditional group, the AI-Assisted group showed significantly shorter surgical turnover time (23.20 ± 2.97 vs. 38.00 ± 2.91 min, P < 0.001) and patient waiting time (5.93 ± 1.13 vs. 11.15 ± 4.24 min, P < 0.001), as well as higher patient satisfaction (98.39% vs. 88.28%, P = 0.018). No significant difference was observed in first-surgery delay rate; surgeon satisfaction was reported descriptively because the session-level ratings were not independent. Multivariable regression analyses adjusting for age, sex, and duration of surgery showed that the AI-Assisted group remained independently associated with these improvements (all P < 0.05 for turnover time, waiting time, and patient satisfaction). In this preliminary single-center study, the LSTM-based model showed encouraging end-of-sequence predictive performance and was associated with improved efficiency in operating room workflows. The reported RMSE does not establish predictive accuracy at earlier intraoperative time points. These findings are exploratory and need to be confirmed in larger, multi-center studies.