Foundation model for gait kinetics estimation across neurological and developmental cohorts from flexible kinematic inputs

In neurological disorders, gait kinetics can inform clinical assessment and treatment planning. Conventional gait-kinetic measurement remains confined to specialised gait laboratories, and the alternative modelling approaches are limited to a specific population cohort, measurement protocol or input configuration, limiting their clinical utility. PathoGaitFM is a foundation model estimating vertical ground reaction force and joint moments across neurological and developmental cohorts from flexible kinematic inputs. Pretrained on healthy participants, PathoGaitFM reached a mean Pearson correlation of 0.811 in cross-validation across four cohorts of participants with cerebral palsy, typical development, stroke or Parkinson’s disease, showing that one shared model tracked kinetic waveforms across heterogeneous cohorts. In five unseen external datasets it reached 0.898 without retraining, showing consistently high accuracy across laboratories and populations. Incomplete inputs lowered correlation by at most 0.041, supporting reuse across joint-angle sets. These findings motivate prospective validation with flexible kinematic inputs in clinical populations.