Mylène Jansen

242 Chapter 12 developed in-house (for details see supplementary file). Bone segmentation was performed semi-automatically, after which the perpendicular distance from the tibia plateau to the femur was measured throughout the entire joint. Only tibial areas where the perpendiculars were ‘reflected’ back onto the tibia surface (i.e. the femoral perpendicular originating from the location where the tibial perpendicular meets the femoral surface, has to meet the tibial surface as well) were included, to only include joint space areas where mutual force transfer between the 2 bones can take place. The medial and lateral boundaries were determined similar as for KIDA evaluation: the width of the medial and lateral side of the joint are 3/20 of the total width of the joint, and the outer border of both sides is 2/15 of the total joint width away from the outer border of the joint, the latter was performed manually (MJ). 31 The median of the remaining perpendicular distances of the MAC was calculated to get the ‘3D CT’ surface median JSW value. The median value instead of the mean value was used to exclude the influence of potentially artificially induced exceptionally large bone-bone distances, however outcome was almost identical in case mean values were used. In addition to the bone-to-bone distance of the 3D image, the coronal CT scans were rotated semi-automatically to a standard position in order to match the position used for the (weight- bearing) radiographs. The tibia plateau was positioned parallel to the axial plane and the line through the back of the femoral condyles was positioned parallel to the coronal plane, viz the most optimal 2D image acquisition. The positioning of tibia in relation to femur was not changed (i.e. no artificial changes were made in the amount of flexion). Subsequently, an over-projection of the repositioned CT scan was created in the coronal plane, so that a non- weight-bearing 2D radiograph was mimicked. A wedge was added based on the current pixel size. These radiographs were then analyzed using the KIDA software, according to the same method and by the same observer as used for the weight-bearing radiographs. The ‘2D CT’ MAC mean JSW was calculated. A representative image of the 4 different techniques for the same patient is shown in Figure 2. Statistical analyses For patient characteristics and image analysis results, descriptive statistics were used. Pearson R correlations were calculated between the techniques cross-sectionally, using all patient time points in 1 comparison. Additionally, Pearson R correlations between the techniques were calculated for the changes over time (2 years - baseline). To describe correlation strength, the guide for R -values suggested by Evans in 1996 was used: <0.2 very weak; 0.2 – 0.39 weak; 0.40 – 0.59 moderate; 0.60 – 0.79 strong; >0.8 very strong. 33 P- values <0.05 were considered statistically significant. IBM SPSS Statistics version 25 (IBM Corp; Armonk, NY) was used for all statistical analyses.