Objective:

To develop a deep learning system for recognizing phases in pediatric laparoscopic fundoplication and assess its phase classification, reconstruction of operative workflow, and estimation of phase duration.

Approach:

• Study design: Retrospective video study of laparoscopic fundoplications performed at the University of Tsukuba Hospital from January 2015 through December 2025.
• Dataset: The final dataset comprised 38 operative videos after exclusions from 50 identified procedures. Video resolution and frame rate varied across cases.
• Annotation: Investigators defined 10 workflow phases and annotated videos in 1-second increments. Two pediatric surgeons performed initial annotation, followed by review of every video by one surgeon; disagreements or unclear boundaries were discussed.

Key Findings:

• The supplied context describes the study design and annotation process but does not report model performance or other study results.

Interpretation:

The authors intended the study to provide an initial benchmark for automated workflow analysis in pediatric minimally invasive surgery and to facilitate subsequent applications in surgical education and performance assessment.

Limitations:

• The supplied context ends during the annotation methods and does not include model architecture, evaluation results, or the remainder of the methods.
• The dataset included 38 videos from a single hospital.

Conclusion:

The supplied excerpt describes the study design and annotation methods but does not report model performance or study results.

Sources:

• Article context provided by the user