P166 - Automated estimation of the mechanical Hip-Knee-Ankle angle using standard knee radiographs.
To develop an automated image analysis pipeline to measure the femoro-tibial angle (FTA) from standard knee radiographs. Furthermore, we aimed to investigate which FTA method correlates best to the mechanical hip-knee-ankle angle, as measured from full-limb radiographs.
Methods and Materials
We included 110 pairs of standard knee radiographs and full-limb radiographs of 100 patients. We used automatic search algorithms to find anatomic landmark points on standard knee radiographs. Based on the landmark points, we automatically calculated the FTA according to nine different methods (six described in the literature and three newly developed) and tested which methods correlated strongest (Pearson, R-squared and intra-class correlation coefficient (ICC)) to the hip-knee-ankle angle (HKAA), as measured on full-limb radiographs. Subsequently, we used the four FTA methods that correlated most strongly to predict the HKAA in a five-fold cross-validation setting.
Across all pairs of images, the Pearson correlations between FTA and HKAA ranged between 0.830 and 0.901. The R-squared values ranged from 0.689 to 0.812, and the ICC values from 0.832 to 0.897. In the cross-validation experiments to predict the HKAA, these statistics only decreased minimally. The mean absolute error for the best method to predict the HKAA from standard knee radiographs was 1.77° (± 1.29° SD).
We showed that the HKAA can be predicted from standard knee radiographs using an automated pipeline with a mean error of 1.77°. The proposed pipeline enables research of the relationship between varus/valgus malalignment and knee pathology (e.g. osteoarthritis) in studies lacking full-limb radiography.