Agreement between subjective evaluations and a markerless AI-based gait analysis system during lungeing assessment in traditional racehorses

Background: Subjective lameness evaluation during lungeing is routinely performed in equine practice, but its consistency remains limited, especially in cases of mild or complex asymmetry. Aims: This study aimed to assess the agreement between subjective gait evaluations and a markerless AI-based gait analysis system (OAI-MS) in traditional racehorses during lungeing. Intra- and inter-observer agreement of subjective evaluations was also investigated. Methods: 24 traditional racehorses were assessed during routine pre-race inspections (T0) while trotting on a soft surface. Two experienced equine clinicians independently evaluated each horse on both reins using the AAEP 0–5 scale; scores were then converted to a 3-level ordinal scale (0 = sound, 1 = mild, 2 = severe). Simultaneously, gait data were collected using the OAI-MS. A subset of 10 horses was re-evaluated after 10 days (T1) to assess short-term repeatability of the OAI-MS. Video-based reassessment (T2) was used to evaluate intra-observer agreement. Agreement was calculated using weighted Cohen's and Fleiss’ kappa. p < 0.05. Results: Inter-observer agreement ranged from κ = –0.20 to 0.36. Agreement between subjective evaluators and the OAI-MS ranged from slight to moderate (κ = 0.13–0.47). Intra-observer agreement was fair (κ ≈ 0.22), and OAI-MS repeatability reached κ = 0.43. Agreement was higher for forelimbs than hindlimbs. Most discrepancies were of low magnitude. Conclusion: Subjective gait evaluations during lungeing showed limited agreement. The OAI-MS demonstrated moderate repeatability, supporting its usability in the field and its potential role as a complementary tool in clinical decision-making, particularly when asymmetries are mild or disagreement occurs.