When conducting motion analysis using 3-dimensional motion capture technology, errors in marker placement on the knee results in a widely observed phenomenon known as “crosstalk” [1-18] in calculated knee joint angles (i.e., flexion-extension (FE), adduction-abduction (AA), internal-external rotation (IE)). Principal Component Analysis (PCA) has recently been proposed as a post hoc method to reduce crosstalk errors and operates by minimizing the correlation between the knee angles [1, 2]. However, recent studies that have used PCA have neither considered exercises, such as cycling (C) and elliptical training (E), other than gait (G) nor estimated the corrected knee axes following PCA correction. The hypothesis of this study is that PCA can correct for crosstalk in G, C, and E exercises but that subject-specific PCA corrected axes differ for these exercises.

Motion analysis of the selected exercises were conducted on 8 normal weight (body mass index (BMI) = 21.70 +/- 3.20) and 7 overweight participants (BMI = 27.45 +/- 2.45). An enhanced Helen Hayes marker set with 27 markers was used to track kinematics. Knee joint FE, AA, and IE angles were obtained with Cortex (Motion Analysis, Santa Rosa, CA) software and corrected using PCA to obtain corrected angles for each exercise. Exercise-specific corrected knee joint axes were determined by finding axes that reproduced the shank and ankle body vectors taken from Cortex when used with the PCA corrected angles. Then, PCA corrected gait axes were used as a common set of axes for all exercises to find corresponding knee angles. Paired t-tests assessed if FE-AA angle correlations changed with PCA. Multivariate Paired Hotelling’s T-Square tests assessed if the PCA corrected knee joint axes were similar between exercises. ANOVA was used to assess if Cortex angles, PCA corrected angles, and knee angles using PCA corrected gait axes were different.

Reduced FE-AA angle correlations existed for G (p<0.001 for Cortex and p=0.85 for PCA corrected), C (p=0.01 for Cortex and p=0.77 for PCA corrected), and E (p<0.001 for Cortex and p=0.77 for PCA corrected). Differences in the PCA corrected knee axes were found between G and C (p<0.0014). Then, differences were found between Cortex, PCA corrected, and C and E knee angles using the PCA corrected G axes (p<0.0056).

The results of this study suggest that if PCA is used to reduce crosstalk errors in motions other than G then it is recommended to adopt the use of a PCA corrected axes set determined from G to produce the PCA corrected angles.