AI Is Learning to Read What Surgeons Feel: Tissue Tension From Surgical Video
- Aug 25
- 2 min read
The 45-Minute Problem
Surgical video is now data.
Most laparoscopic and robotic procedures are recorded by default.
Reviewing that footage is a bottleneck.
An experienced surgeon needs about 45 minutes to review a single case, according to Dr. Frans van Workum, a gastrointestinal surgeon at Radboud University Medical Centre and Canisius Wilhelmina Hospital.
The Netherlands alone generates more than 100,000 hours of surgical video every year.
No workforce can watch it all.
What Current Models Already See
Machine-learning systems already score several parts of a procedure.
They track how long each surgical step takes.
They count unnecessary instrument movements.
They estimate blood loss.
On these metrics, algorithmic assessments already track closely with those of experienced surgeons.
The Missing Signal: Tissue Tension
One variable has resisted automation: tissue tension.
Surgeons stretch and separate tissue to perform precise dissection.
Reading how much tension a tissue can bear, and where it concentrates, is what keeps a dissection from damaging the structures around it.
Van Workum describes it as instinct built over years, not something that shows up on a monitor.
Computer vision systems cannot sense it.
Training on 35 Rectal Cancer Cases
A team from HealthTech Nexus, the joint research program between the University of Twente and Radboudumc, is testing whether that instinct can be learned from video alone.
Thirty-five surgeons will contribute footage of rectal cancer surgeries for the study.
The models will be trained to read three signals: how far tissue stretches, which instruments are in use, and how those instruments move across each phase of the procedure.
Dr. Estefanía Talavera Martínez, assistant professor at the University of Twente, is leading the algorithm-training side of the project.
The benchmark is direct: can the system estimate tension as reliably as an experienced surgeon does.
Why Transparency Is the Point
The team chose white-box models over black-box ones.
That decision is deliberate, not incidental.
A white-box model exposes its own reasoning, so a surgeon can see why the system reached a given estimate and judge whether that reasoning holds up.
A black-box confidence score offers no such check.
For a signal this safety-critical, that difference matters more than raw accuracy.
Early Stage, Not Proven
The project is in an early phase.
No results exist showing whether the models can match surgeon-level tension estimates.
If the approach works, the payoff is time: less of the 45 minutes per video spent on manual review, and an objective basis for comparing surgical techniques.
Van Workum frames the ambition modestly: feedback that improves surgical performance by even five percent, he says, would meaningfully change everyday practice and patient outcomes.
Whether AI can actually supply that feedback remains open.
