Workflow understanding
Phase, step, action, and event recognition that turn long recordings into temporal structure.
Surgical video · Public research
Understand what can be learned from surgical video.
Surgical video can support many different AI questions. The key is knowing the task, the annotation, the timing, and the evidence standard before comparing models or datasets.
Task map
Start with the question, then choose data, labels, and metrics.
A single operation recording can become a workflow timeline, a segmentation benchmark, a safety review dataset, or a training-feedback tool. Those are not interchangeable tasks.
Scene understanding
Detection, segmentation, depth, and anatomy-aware labeling that describe what is visible.
Performance and quality signals
Skill, safety criteria, economy of motion, and review tasks that require careful clinical definitions.
Annotation and benchmarking
Label design, inter-rater disagreement, leakage checks, and benchmark choices that shape trust.
What video can support
The same recording can serve very different research questions.
Task 01
Procedure workflow
Turn long procedures into structured timelines for research and review.
- AI task
- Phase recognition
- Annotation
- Temporal phase labels
Task 02
Instrument activity
Use visible instruments as signals for activity, workflow, and interaction.
- AI task
- Detection and tracking
- Annotation
- Boxes, classes, and trajectories
Task 03
Surgical scene anatomy
Represent anatomy and instruments spatially for scene-understanding tasks.
- AI task
- Detection and segmentation
- Annotation
- Boxes and pixel-level masks
Task 04
Safety-criteria review
Translate clinical definitions into explicit, reviewable dataset targets.
- AI task
- Criteria assessment
- Annotation
- Frame- and criterion-level labels
Go deeper
Compare annotations, access, and source papers.
The Weekly Signal
Get the weekly Surgical AI signal.
A concise email with practical explainers, useful papers, dataset updates, tools, and ways to participate.