Background/Objective: Consistent determination of the anatomical landmarks on image or image-based three dimensional (3D) models is a basic requirement for reliable analysis of the human joint kinematics using imaging techniques. We examined the intra- and inter-observer reliability of determination of the medial and lateral epicondyle landmarks on 2D MR images and 3D MRI-based models of the knee. Materials and Methods: Sixteen coronal plane MRI recordings were taken from 18 healthy knees using a knee coil with T2-weighted fast spin-echo sequence and 512×512 pixel size. They were then processed by the Mimics software to provide the coronal and axial plane views and to create a 3D image-based model of the femur. Each image was reviewed twice, at least one-day apart. The interclass correlation coefficient, standard error of measurement, and coefficient of variation were calculated to assess the intra- and inter-observer reliability of the landmark determination by six experienced radiologists. A mixed model analysis of variance (ANOVA) with two days of observation as the within-subject factor, and observers (six radiologists) and methods (2D vs. 3D) as between-subject factors were used to test the effect of observer, two days of observation and method of evaluation on landmark determination. Results: The results indicated that the interclass correlation coefficients for the intraobserver and inter-observer determination of landmarks on images and image-based 3D models were above 0.97. The standard error of measurement ranged between 0.41 and 0.78 mm for x; 1.35 and 3.43 mm for y; and 1.03 and 4.71 mm for z coordinates. Furthermore, the results showed no significant difference for within and between-subject comparisons of each coordinate of the lateral epicondyle as well as x and z coordinates of the medial epicondyle. For the y coordinate of the medial epicondyle, the p value of within-subject comparison was borderlinely significant (p=0.049).
Conclusion: It was concluded that the intra- and inter-observer reliability of the bony landmark determination on both image and image-based 3D models were excellent.