Objective:

To develop and assess an automated phase-recognition model for robotic choledochal cyst surgery and to evaluate its clinical utility in understanding workflow dynamics.

Approach:

• Study Design: A retrospective, single-institution observational study was conducted, consisting of two components: developing an automated model for surgical phase recognition and applying it to an independent clinical cohort.
• Surgical Phase Definition: Nine clinically meaningful phases of robotic choledochal cyst surgery were predefined, including preparation, bile-duct dissection, ductoplasty, hepaticojejunostomy, and others.
• Model Development: An automated model for surgical phase recognition was developed using videos from robotic surgeries, with a board-certified pediatric surgeon annotating the phases.

Key Findings:

• The automated phase-recognition model was developed and internally evaluated.
• Phase durations derived from AI could provide clinically meaningful workflow information beyond total operative time.
• Longer dissection phases were anticipated to correlate with intraoperative burden.

Interpretation:

Automated workflow measures could assist in understanding procedure-specific workflow differences and variability patterns at the phase level.

Limitations:

• The study was retrospective and conducted at a single institution.
• Formal interobserver agreement for phase annotation was not evaluated.

Conclusion:

The investigation demonstrates the potential of AI in quantifying workflow variation in robotic surgery, particularly in complex procedures like choledochal cyst excision.