The Missing Knowledge Layer
for Surgical AI
SurgicalDataOS transforms cataract surgery videos into structured machine knowledge that powers surgical AI, robotics, simulation, research and next-generation ophthalmic intelligence.
A knowledge representation framework for machine-understandable cataract surgery.
Knowledge Object Explorer
Knowledge Object Explorer
See how SurgicalDataOS transforms a single surgical event into structured machine-readable knowledge.
Full procedure reference video
Title
Phaco Fixation Established
Representative Frame
40
Frame Range
0–82
Timestamp Range
0.00–1.37 seconds
Phase
Nucleus Management
Stage
Fragmentation
Observation
The phaco tip is embedded within the nucleus, providing stable fixation. The nucleus remains intact, and the chopper has not yet initiated the chopping manoeuvre.
Interpretation
Stable fixation has been achieved, creating the mechanical conditions required for controlled advancement of the chopper toward the equator.
Decision
Maintain secure nuclear fixation while preparing to position the second instrument for nucleus fragmentation.
Action
The phaco tip maintains stable purchase on the nucleus while fixation is preserved.
Event
A stable mechanical relationship is established between the nucleus and the phaco tip, preparing the nucleus for controlled fragmentation.
Outcome
The nucleus is securely stabilised, allowing safe progression to the chopping phase.
Cognitive Intent
Create a stable mechanical foundation from which a controlled and reproducible nuclear fracture can be initiated.
Knowledge Extract
Successful vertical chop begins with secure nuclear fixation. Stable fixation is the prerequisite that enables controlled force transmission during the subsequent chopping manoeuvre.
Dataset Information
Title
Nucleus Fragmentation
Procedure
Phacoemulsification
Technique
Vertical Chop
Duration
13.7
FPS
60
Current Knowledge Object Number
Knowledge Object 1 of 7
Knowledge Graph
Coming in next implementation.
The Problem
Current Surgical AI is trained on pixels.
Not on surgical knowledge.
Object detection and segmentation tell a model where things appear in a frame. They do not encode why an action occurs, when a phase transitions, or how instrument motion relates to tissue response. Robotic cataract surgery demands causal, temporal, and intent-aware representations — not bounding boxes alone.
- Detects objects in isolated frames
- No temporal or causal structure
- Cannot reason about surgical intent
- Fails under domain shift and occlusion
- Structured representation of surgical semantics
- Temporal, causal, and intent-aware reasoning
- Enables robotic planning and skill assessment
- Generalizes across surgeons and equipment
The Knowledge Model
A representation framework for machine-understandable surgery
SurgicalDataOS defines a formal ontology that transforms raw surgical video into a queryable knowledge graph — connecting perception, action, anatomy, and outcome at every level of abstraction.
Video
Continuous surgical recording as the root signal source.
Frame
Atomic temporal unit indexed for precise event alignment.
Event
Time-bounded occurrence with defined onset and offset.
Action
Deliberate surgical maneuver performed by the operator.
Instrument
Tool identity, pose trajectory, and interaction state.
Anatomy
Ocular structures with spatial relationships and geometry.
Intent
Clinical objective driving the current surgical decision.
Decision
Branch point where the surgeon selects among alternatives.
Outcome
Measured result linking action sequence to patient state.
Knowledge Graph
All entities connect through typed edges — enabling traversal from any frame to its surgical context, downstream outcomes, and training signal for AI and robotic systems.
Representation Standards
Hierarchical annotation framework
Every annotation in SurgicalDataOS maps to a defined level in our taxonomy — from full procedure context down to sub-millimeter tissue geometry. This hierarchy ensures consistency, composability, and machine-readability across datasets and institutions.
Beyond Computer Vision
Existing datasets describe what is visible.
SurgicalDataOS represents what is happening.
Computer vision identifies objects in individual frames. Surgical intelligence requires understanding temporal events, anatomical relationships, instrument interactions, surgical intent and procedural decision making. SurgicalDataOS transforms surgical video into machine-understandable knowledge.
Traditional Computer Vision
Visual Perception
Where is it?
- Detection
- Segmentation
- Tracking
- Classification
- Bounding Boxes
Describes pixels and objects within individual frames.
Knowledge Objects
Surgical Understanding
What is happening?
- Phase
- Action
- Instrument
- Anatomy
- Tissue
- Complication
Represents surgical workflow, anatomical context and procedural meaning.
Knowledge Layer
Machine Reasoning
What should happen next?
- Knowledge Graph
- Decision Layer
- Surgical Context
- Robot-ready Representation
- Foundation Model Input
Structured knowledge that enables reasoning, planning and robotic execution.
Applications
One knowledge layer, many frontiers
Research
Accelerate hypothesis testing with queryable surgical knowledge graphs instead of raw video archives.
Robotic Surgery
Train manipulation policies on structured action-outcome pairs with temporal and causal alignment.
Foundation Models
Pre-train vision-language models on semantically rich surgical narratives, not pixel co-occurrence.
Simulation
Drive physics-informed simulators with real procedure dynamics, instrument trajectories, and tissue response.
Skill Assessment
Quantify surgical proficiency through decision trees, complication rates, and micro-action efficiency.
Autonomous Workflow
Enable phase-aware automation that understands context, not just detects objects in frame.
Clinical Decision Support
Surface evidence from structured outcome data to inform intraoperative and postoperative decisions.
Platform
Infrastructure for surgical knowledge at scale
Annotation Studio
Multi-level labeling with AI assist, consensus workflows, and real-time validation against the representation schema.
Dataset Marketplace
Discover, license, and version curated cataract datasets with full provenance and quality metrics.
Knowledge Graph Explorer
Traverse surgical entities, query temporal relationships, and export subgraphs for model training.
Validation Dashboard
Inter-annotator agreement, schema compliance, and automated quality gates before dataset release.
API
Programmatic access to annotations, graph queries, and streaming video pipelines for research integration.
About
“The knowledge layer for machine-understandable surgery.”
SurgicalDataOS is founded on a simple conviction: the next generation of surgical AI will not emerge from larger models trained on more pixels. It will emerge from structured representations of surgical expertise that preserve observation, reasoning, decision-making and action in a machine-understandable form.
From Surgical Video to Machine Knowledge
From Surgical Video to Machine Knowledge
Every surgical procedure is represented as a sequence of Machine Knowledge Objects (MKOs)—structured computational representations that preserve what the surgeon observed, interpreted, decided and performed. Together these MKOs form a machine-readable Knowledge Graph capable of supporting explainable AI, surgical robotics, simulation, education and collaborative research.
MKO-0001
Phase
Nucleus Fragmentation
Observation
Stable phaco fixation established.
Interpretation
The nucleus is securely stabilised, permitting controlled transmission of chopping forces.
Decision
Advance the chopper towards the equator.
Action
Controlled chopper advancement.
MKO-0002
Phase
Primary Nuclear Fracture
Observation
Opposing instrument forces create a central crack.
Interpretation
Mechanical stress exceeds nuclear cohesion.
Decision
Complete the fracture while maintaining chamber stability.
Action
Vertical chop executed.
MKO-0003
Phase
Quadrant Removal
Observation
Fragment engaged under stable occlusion.
Interpretation
Continuous aspiration allows controlled centralisation.
Decision
Maintain occlusion while repositioning the fragment.
Action
Fragment centralised and emulsified.
Beyond Annotation
Beyond Annotation
Current datasets describe pixels.
SurgicalDataOS represents surgical intelligence.
- Labels pixels
- Detects instruments
- Segments anatomy
- Describes events
- Stores video metadata
- Creates Machine Knowledge Objects
- Captures surgical reasoning
- Represents operative intent
- Builds Knowledge Graphs
- Enables robotic decision making
Research Initiative
Research Initiative
SurgicalDataOS is an independent research initiative led by Dr. Merine Paul, an ophthalmic surgeon with more than twenty-five years of experience in cataract surgery and surgical education.
The project emerged from a simple observation: surgical videos contain substantially more knowledge than today's AI systems are able to learn.
Its objective is to create a structured knowledge representation for surgery that enables explainable AI, surgical robotics, simulation, education and collaborative research.
Collaboration
Building Surgical Intelligence Together
SurgicalDataOS is an open research initiative that welcomes collaboration with surgeons, AI researchers, robotics companies, academic laboratories and industry partners interested in advancing machine-understandable surgical intelligence. We believe the future of surgical AI will be built through open scientific collaboration, shared knowledge representations and rigorous validation.
Contact
Build the knowledge layer with us
Whether you are advancing surgical AI research, developing robotic platforms, or curating clinical datasets — we want to hear from you.