Effect of the conventional gait model 2 variants on lower-limb kinematics in individuals with cerebral palsy

Three-dimensional gait analysis is widely used to support clinical decision-making in individuals with motor impairments. However, kinematic outputs depend strongly on the underlying biomechanical model. The open-source Conventional Gait Model II (CGM2) integrates updates to joint centre estimation (CGM2.1), inverse kinematics (CGM2.2), and cluster-based segment tracking (CGM2.3). While previous work demonstrated consistency among CGM2 variants in typically developing children, their effect in clinical populations remains unknown. This study quantified how CGM2 variants influence gait kinematics in individuals with cerebral palsy (CP).

Twenty-one individuals with CP (GMFCS I-II) underwent gait analysis using a 12-camera motion capture system and a CGM2.3 marker set. Hip, knee, and ankle kinematics from 487 gait cycles were computed using pyCGM2. Differences between CGM2.1, CGM2.2, and CGM2.3 were evaluated using Mean Absolute Deviation (MAD) and the adjusted coefficient of determination (R²).

Overall, small differences were observed between model variants. MAD values were typically below 5° for most joints and planes, with high correlation between curves (R²>0.7). Hip rotation showed the largest discrepancies, with maximum MAD up to 7.7° when comparing CGM2.2 and CGM2.3. Differences between CGM2.1 and CGM2.3 were greater in the transverse and frontal planes but remained within acceptable limits (<5°), except for hip rotation.

The CGM2 variant selection has limited impact on gait kinematics in individuals with CP, and most differences fall within known repeatability error. However, transverse-plane kinematics, particularly hip rotation, should be interpreted with caution when comparing data across CGM2 variants.