Validation of a handheld smartphone markerless method for quantifying thoracolumbar flexion-extension range of motion in horses at trot.

Overview

• This study evaluated the accuracy of a handheld smartphone-based system (RealHorse®) for measuring back movement in horses trotting, by comparing it to a gold-standard optical motion capture system (Qualisys®).
• The research found that the smartphone method showed good agreement with the reference system, especially when averaging measurements over multiple strides, indicating its potential use for objective, field-based assessment of horse back motion.

Background and Purpose

• Back pain and poor performance in horses are often difficult to assess objectively due to the limitations of subjective observation.
• The thoracolumbar flexion-extension (FE) range of motion is an important indicator of spinal health, but current methods to quantify it are either subjective or require expensive equipment.
• The study investigates a handheld markerless computer vision system called RealHorse® (RH), which uses a smartphone to record and analyze horse back movement without markers—making it potentially practical for clinical and field use.
• The objective was to validate RH by comparing its FE measurements with those from a well-established optical motion capture system, Qualisys® (QS), which serves as the reference standard.

Methods

• Population: 59 horses were recorded trotting on a straight line and a subset of 23 were also recorded trotting on a circle.
• Data Collection:
  • Both RH and QS recorded simultaneously during trials to capture thoracolumbar FE motion.
  • A light source was used to manually synchronize frames between the two systems.
  • The RH algorithm had not been trained on these validation videos, ensuring unbiased testing.
• Analysis:
  • Agreement between RH and QS was assessed separately for straight-line and circular trot conditions.
  • Measurements were analyzed at two levels:
    • Stride-level – comparing each individual stride’s measurements.
    • Trial-level – averaging measurements over multiple strides within a trial.
    • Metrics used were:
      • Mean Signed Error (MSE) – measures average direction and magnitude of bias.
      • Mean Absolute Error (MAE) – measures average magnitude of error, regardless of direction.
      • Bland-Altman Limits of Agreement (LoA) – defines the range within which most differences between methods fall.

Results

• Straight-line Trot:
  • Stride-level results:
    • MSE: -0.08°, nearly zero indicating minimal bias.
    • MAE: 0.96°, indicating small average difference between RH and QS for individual strides.
    • LoA: from -2.49° to 2.32°, meaning most stride-to-stride measurement differences were within approximately ±2.5°.
  • Trial-level results (averaging strides):
    • MSE: -0.13°, still minimal bias.
    • MAE: 0.44°, smaller error than stride-level, showing that averaging reduces variability.
    • LoA: -1.22° to 0.94°, indicating tighter agreement than stride-level.
• Circle Trot:
  • Stride-level results:
    • MSE: -0.71°, slightly more bias than straight-line.
    • MAE: 1.14°, indicating somewhat larger average errors.
    • LoA: -3.17° to 1.74°, a wider range, indicating greater variability.
  • Trial-level results:
    • MSE: -0.62°, moderate bias.
    • MAE: 0.78°, errors less than stride-level but larger than straight-line trial-level.
    • LoA: -2.18° to 0.93°, showing moderate agreement.
• Key observations:
  • Agreement was better when horses trotted on a straight line compared to a circle due to likely increased biomechanical variability on circles.
  • Averaging measurements across strides (trial-level) consistently improved agreement compared to stride-by-stride analysis.

Conclusions and Implications

• The RealHorse® smartphone system can provide objective, markerless measurement of thoracolumbar flexion-extension range of motion in horses trotting, with accuracy close to a gold-standard motion capture system.
• It is particularly suitable for repeated monitoring within the same horse, as averaging over strides reduces measurement noise.
• This method may support clinical and field evaluations of back mobility, aiding in diagnosis and monitoring of back pain or performance issues.
• However, clear clinical decision thresholds to interpret the measurements from RealHorse® still need to be established before widespread clinical adoption.
• Stride-to-stride variability inherent in horse movement affects measurement precision, so multiple strides should be analyzed rather than relying on single stride data.