Variability, agreement and reliability of MRI knee landmarks

Surface mesh reconstructions of bones are often required to define landmark-based coordinate systems, regions of interest and morphological features when studying the soft tissues of the knee from MRI scans. This study reports the variability, agreement and reliability of osseous landmarks to better understand their downstream effects. Fifteen landmarks were defined on the distal femur and twelve on the proximal tibia. Surface meshes were created from twenty right knee MRI scans with a mean subject age of 30.9 years. A single observer identified landmarks on all twenty knees, while three observers repeated the observations three times on a subset of eight knees. All observations were aligned to the Procrustes mean shapes. Principal component analysis was used to study inter-subject variability and two-way ANOVA for inter- and intra-observer agreement and reliability. Inter-subject landmark variation ranged from 0.6 to 5.26 mm, while inter- and intra-observer agreement were at most 5.1 and 5.69 mm respectively. Between-observer reliability ranged from 0.07 to 0.98 while within-observer values were between 0.51 and 0.98. Landmarks derived from fitted spheres or circles often performed well, while most others had their poorest agreement or greatest variation limited to only one or two cardinal directions.