ROBUST-MIPS: A Combined Skeletal Pose and Instance Segmentation Dataset for Laparoscopic Surgical Instruments.

Area of Science:

• Computer-assisted interventions
• Medical imaging analysis
• Machine learning for surgery

Background:

• Accurate surgical tool localization is crucial for computer-assisted interventional technologies.
• Deep learning models for segmentation tasks are limited by the availability of diverse annotated data.
• Current annotation methods for surgical tools are time-consuming and data-intensive.

Purpose of the Study:

• To propose skeletal pose annotations as a more efficient alternative for labeling surgical tools.
• To introduce the ROBUST-MIPS dataset, combining tool pose and instance segmentation.
• To facilitate the joint study and comparison of pose and instance segmentation annotation styles.

Main Methods:

• Developed the ROBUST-MIPS dataset by enriching the existing ROBUST-MIS dataset with skeletal pose annotations.
• Established a benchmark using popular pose estimation methods to evaluate the efficacy of pose annotations.
• Released benchmark models and custom annotation software to promote adoption.

Main Results:

• Demonstrated high-quality results using skeletal pose annotations for surgical tool localization via a benchmark study.
• The ROBUST-MIPS dataset enables head-to-head comparisons of different annotation styles on downstream tasks.
• Pose annotations provide a balance between semantic richness and annotation efficiency.

Conclusions:

• Skeletal pose annotations are an effective and efficient method for surgical tool localization in AI.
• The ROBUST-MIPS dataset and associated tools lower the barrier for research in this area.
• This work accelerates the development of advanced computer-assisted surgical technologies.