Prediction of remaining surgery duration based on machine learning methods and laparoscopic annotation data.

Area of Science:

• Medical Informatics
• Surgical Technology
• Machine Learning in Medicine

Background:

• Operating room efficiency is crucial for surgical care quality.
• Accurate prediction of surgery duration aids scheduling and resource management.
• Laparoscopic cholecystectomy is a common procedure where duration prediction is beneficial.

Purpose of the Study:

• To develop and evaluate a semi-automated machine learning method for predicting remaining surgery duration (RSD).
• To optimize operating room scheduling and resource allocation through improved surgery duration prediction.
• To enhance the quality of surgical care by providing real-time RSD estimates.

Main Methods:

• A Random Forest regression model was implemented using the Cholec80 dataset.
• The model utilized two data streams: surgical phase and tool type at each time-frame.
• Performance was evaluated using Mean Absolute Error (MAE) on training, validation, and test sets.

Main Results:

• The machine learning approach achieved a Mean Absolute Error (MAE) of 5.89 minutes for overall surgery duration.
• A more precise MAE of 4.61 minutes was obtained when predicting 20 minutes prior to the operation's end.
• The model demonstrated effective prediction of remaining surgery duration in laparoscopic cholecystectomy.

Conclusions:

• Employing two distinct regression models, switched at an elapsed time threshold, significantly improves RSD prediction.
• This approach offers enhanced accuracy compared to methods solely processing endoscopic video.
• The developed method provides a valuable tool for real-time surgical duration estimation and operating room management.