PhysioMatrix — Open Research Platform

Turning expert judgment into reproducible research data

Two physiotherapists can watch the same squat and reach different conclusions - and there is usually no record of what they actually judged, or how much they agreed. PhysioMatrix gives clinical researchers a structured way to record, score, and measure agreement on movement quality, so a rubric can be validated instead of just trusted.

Front
Side
Mid-right
Mid-left
What is PhysioMatrix?

A web platform where physiotherapists rate recorded exercises, so we can measure how consistent expert assessment really is

Study participants perform standardized exercises (squats, stabilization tests, back exercises, DNS tests) in a motion lab while four cameras record them. Registered physiotherapists then watch these recordings online and rate each exercise against a fixed set of criteria. Because many experts rate the same recordings, we can calculate how often they agree, which criteria are reliable, and where the rubric needs work.

For physiotherapists

Become an evaluator

  • Register, and an administrator verifies your account.
  • Sign in and you get one exercise at a time with its instruction video and camera views.
  • Rate each criterion on a 1-5 scale. You can stop at any time and continue later.
  • Your dashboard shows how many exercises you have rated.
For researchers

Validate an assessment rubric

  • Define exercise types, their criteria and instruction videos.
  • Follow agreement between evaluators per exercise and per criterion.
  • Run the reliability study and export the scored data.
  • Compare expert ratings with computer-vision pose measures.
About the data

Participant privacy

  • Faces are blurred before videos are uploaded to the platform.
  • Recordings are only streamed to signed-in, verified evaluators.
  • Evaluators never see each other's scores before submitting their own.
How it works
1
Capture from four angles

Every exercise is recorded simultaneously from front, side, mid-left, and mid-right cameras, so a reviewer never has to guess what a bad viewing angle hid.

2
Score against a shared rubric

Evaluators rate each defined criterion on the same 1-5 scale, independently of one another - no evaluator sees another's scores before submitting their own.

3
Measure whether experts actually agree

Every exercise reviewed by more than one evaluator gets a real reliability score: mean pairwise agreement and a linear-weighted Cohen's Kappa per criterion, not just a review count.

4
Publish an open, reproducible dataset

The resulting scored, multi-angle dataset and the platform itself are released under Apache 2.0 - built to be checked, reused, and extended, not just looked at.

1-5shared rubric scale, defined per exercise criterion
2independent reliability metrics: agreement % and weighted Cohen's Kappa
4synchronized camera angles per recorded exercise
Apache 2.0platform and dataset tooling, open source
Developed at

Built at CIIRC CTU (Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague) by Jindřich Adolf.

Clinical research partner

In cooperation with the 2nd Faculty of Medicine, Charles University, represented by Václav Pecka.