Related Experiment Video
Updated: Jul 4, 2025
06:18
Author Spotlight: Segmentation and VR for Advanced Neurovascular Interventions
Published on: April 5, 2024
1.0K
Surgical phase and instrument recognition: how to identify appropriate dataset splits
Georgii Kostiuchik1,2, Lalith Sharan3,4, Benedikt Mayer5
1Department of Cardiac Surgery, Heidelberg University Hospital, Heidelberg, Germany. georgii.kostiuchik@med.uni-heidelberg.de.
International Journal of Computer Assisted Radiology and Surgery
|January 29, 2024
Summary
This study introduces a tool to visualize surgical dataset splits, improving machine learning model evaluation. It helps identify and fix data imbalances for better surgical phase and instrument recognition.
Area of Science:
- Medical image analysis
- Machine learning in surgery
- Data science for healthcare
Background:
- Machine learning model evaluation requires representative data splits.
- Surgical workflow and instrument recognition face data imbalance challenges.
- Existing dataset splits often neglect instrument co-occurrence.
Purpose of the Study:
- To present a data visualization tool for exploring surgical dataset partitions.
- To enable interactive assessment of dataset splits for phase and instrument recognition.
- To identify and address sub-optimal dataset splits in surgical machine learning.
Main Methods:
- Developed a publicly available, interactive data visualization tool.
- Focused on visualizing phase occurrences, transitions, and instrument combinations.
- Facilitated the assessment of dataset splits for surgical recognition tasks.
Main Results:
- Analyzed Cholec80, CATARACTS, CaDIS, M2CAI-workflow, and M2CAI-tool datasets.
- Uncovered underrepresented phase transitions and instrument combinations in dataset splits.
- Identified improvements for dataset splits and confirmed tool usability via a user study.
Conclusions:
- Careful dataset splitting is crucial for reliable machine learning evaluation in surgery.
- The interactive tool aids in determining better dataset splits for improved practice.
- The tool is available online to enhance surgical machine learning development.
