LapEx: A new multimodal dataset for context recognition and practice assessment in laparoscopic surgery

Area of Science:

• Surgical Data Science
• Machine Learning in Medicine
• Laparoscopic Surgery

Background:

• Increasing demand for large, realistic annotated datasets in Surgical Data Science (SDS).
• Existing laparoscopic surgery datasets often lack clinical realism and detailed annotations.
• Annotation variability is rarely assessed in surgical datasets.

Purpose of the Study:

• To create a novel, multi-level annotated dataset for laparoscopic sleeve gastrectomy procedures.
• To address the limitations of current datasets in terms of granularity and realism.
• To provide a baseline for evaluating machine learning models in SDS.

Main Methods:

• Compiled 30 sleeve gastrectomy procedures.
• Performed three levels of annotation: fine-grained procedural activities, semantic segmentation, and surgical skill assessment (scene exposition quality).
• Conducted a comprehensive annotation variability analysis.

Main Results:

• Developed a dataset with detailed annotations for fundus dissection in sleeve gastrectomy.
• Quantified annotation variability, highlighting task complexity.
• Established a public dataset for advancing SDS research.

Conclusions:

• The new dataset enhances realism and granularity for machine learning in laparoscopic surgery.
• Annotation variability analysis provides crucial insights for model development.
• The publicly available dataset is a valuable resource for the Surgical Data Science community.