Serial bilateral Posterior Anterior (PA) conventional radiographs were obtained 36 months apart in persons with mild to severe knee OA participating in the Health ABC study, a community based, multi-center cohort study of 3075 white and black men and women aged 70–79 at enrollment. All subjects had risk factors for OA, although many had no radiographic evidence as defined by the KL score. KL scoring was performed by Dr. Hunter. More details about the Health ABC Study and the subject population can be found in a separate publication20
. We randomly selected a sample of 136 (272 knee pairs) participants who underwent a bilateral knee radiographs at both the baseline and follow-up (36 month) time points using a fixed flexion subject positioning protocol17
Any knee with a total joint replacement (1 knee), and any joint that was independently assessed to have lateral compartment OA (54 knees) were excluded from the analysis. Determination of lateral compartment OA was based on a non-zero joint space narrowing (JSN) score in the lateral compartment. The study sample was 51% female, 41% African American, with an average age of 74.3 ± 2.8 years (mean ± standard deviation). The baseline KL scores of the eligible 217 knee baseline follow-up pairs (from 118 subjects) were distributed as follows: 93 (KL = 0), 38 (KL = 1), 13 (KL = 2), 55 (KL = 3), and 18 (KL = 4). The radiography protocol called for an extremity detail film cassette (Agfa Ortho Fine), with a 122 cm film to focus distance. The radiography technique was 70 kVp with a variable mAs (6–16 mAs). The radiographic films were digitized using a Vidar (Herndon, VA) film digitizer with a pixel spacing of 0.085 mm and transferred to a personal computer for analysis.
We have previously developed and documented a software tool written using the C programming language16,19
to delineate the tibiofemoral joint on digitized knee radiographs and enable measurements of JSW(x
) and mJSW. A previous study established the reproducibility of the technique by measuring the root mean square standard deviation (RMSSD) as 0.16 mm (normal knees) and 0.18 mm (OA knees) for mJSW16
measured on duplicate radiographs. The reproducibility study used the software in a fully automated mode, with no manual correction, however, to improve the value of the software tool we also developed an accompanying graphical user interface using the Interactive Data Language (IDL), (Kodak Inc., Boulder, CO) application. A human user operates the tool on a cropped image of the knee joint, examines the software-generated contours that delineated the joint, and can then use semi-automated editing tools to verify and correct the software-drawn margins where necessary16
mJSW was measured as the minimum distance between the delineated femur and tibia margins in the medial compartment. Baseline and follow-up images were displayed as pairs with the reader (GN) blinded to the chronological sequence. Image files were prepared by a different researcher (JD), and randomly assigned filenames ensured that the reader had no knowledge of the correct time sequence. The reader judgment was employed to correct any improper software delineated joint margins and to provide a consistent mJSW measurement from baseline to follow-up for cases such as the examples in .
To enable consistent measures of JSW(x), a coordinate system was created based on anatomical landmarks (). The x-axis, defined as the line tangent to both femoral epicondyles, was placed automatically by the software. The x variable represented the position of the JSW(x) measurement along the projected surface of the joint. The y-axis was placed manually as a line perpendicular to the x-axis and tangent to the greatest prominence of the medial epicondyle. The line x = 1, was defined as the tangent to the greatest prominence of the lateral epicondyle of the knee. The variable x is a dimensionless quantity which can be considered to approximately represent the fraction of the total width of the femoral condyle. The use of this variable and the coordinate system is a potential strength of our method since the JSW measurement location can be reproducibly and consistently determined for different subjects and for different visits of the same patient assuming perfectly consistent knee positioning. In principle, x is independent of knee size, pixel spacing, magnification, and other factors that affect the location along the joint. In practice, x can vary due to knee rotation and patient positioning; one goal of our study was to determine whether these effects reduced the responsiveness of the technique.
Landmarks and definition of coordinate system.
The software displayed cropped images of the epicondyles for both visits simultaneously so that the reader could verify consistent landmark placement for both baseline and follow-up images. All images of the knee joint were placed in a consistent orientation with the medial compartment on the left (x < 0.5) and the lateral compartment on the right (x > 0.5). Measurements of medial compartment mJSW and JSW(x) at seven fixed locations (x = 0.125, x = 0.15, x = 0.175, x = 0.2, x = 0.225, x = 0.25, and x = 0.275), were made using the software tool described above (). JSW(x) was defined as the distance between the femur and tibia joint margins in a direction parallel to the y-axis of our coordinate system.
Location specific measurement of JSW, JSW(x), (x = 0.125–0.275), in the medial compartment. Measurements of JSW(x) are made at the x-locations defined by the coordinate system.
We compared medial compartment JSW(x) to mJSW. As a metric to quantify performance we used the SRM, or the ratio of the mean loss to the standard deviation of the loss. We report the SRM values along with the average baseline JSW, the average change, and the standard deviation for each JSW measure.