Reproducibility of computer-assisted joint alignment measurement in OA knee radiographs

(1) To investigate the reproducibility of computer-assisted measurements of knee alignment angle (KA) from digitized radiographs of osteoarthritis (OA) participants requiring total knee arthroplasty (TKA) and (2) to determine whether landmark choice affects the precision of KA measurements on radiographs.

Using a custom algorithm, femoral, central, and tibial measurement-guiding rules were interactively placed on digitized posteroanterior fixed-flexion knee radiographs by mouse control and positioned according to different anatomic landmarks. The angle subtended by lines connecting these guiding rules was measured by three readers to assess interobserver, intraobserver and experience-inexperience reproducibility. Test-retest reproducibility was evaluated with duplicate radiographs from a healthy cohort. Reproducibility was assessed using root-mean square coefficients of variation (RMSCV%). The Bland-Altman method was performed on data obtained from varying anatomic landmarks (confidence interval, CI= 95%).

From 16 healthy and 30 TKA participants, reproducibility analyses revealed a high degree of intraobserver (n=38, RMSCV=0.56%), interobserver (n=38, RMSCV=0.72%), test-retest (n=16, RMSCV=0.87%) and experience-inexperience (n=38, RMSCV=0.73%) reproducibility with variances below 1%. Varying the orientation of tibial and femoral rules according to anatomic landmarks produced a difference that exceeded an a priori limit of agreement of -1.11 degrees to +1.67 degrees.

Our custom-designed software provides a robust method for measuring KAs within digitized knee radiographs. Although test-retest analyses were only performed in a healthy cohort, we anticipate a similar degree of reproducibility in an OA sample. A standardized set of anatomic landmarks employed for KA measurement is recommended since arbitrary selection of landmarks resulted in imprecise KA measurement even with a computer-assisted technique.