The Effect of Radiographic Positioning on Active Appearance Modelling of Bilateral Knee Osteoarthritis in the Osteoarthritis Initiative

Research output: Contribution to journalAbstractpeer-review


Purpose: To investigate the effect of different radiographic equipment in an Active Appearance Model (AAM) of bilateral knee OA.
Dataset: The male subjects were selected from Image B imaging sub-group of the Osteoarthritis Initiative (OAI), who are all part of the progression cohort. They were age, height and weight matched to subjects from the incidence cohort, after initial selection for beam angle of 10° (±2.5°).
Methods: The new bilateral knee model was based on single knee active shape models from previous studies. A template of 186 points was fitted to each image using the AAM software to describe the shape of the knees. The co-ordinates of the points were found semi-automatically in all images and principal components analysis (PCA) was used to reduce these variables to a smaller number of independent variables (modes). The distance between the knees was measured using Image J and actual distance (mm) was calculated using the pixel resolution of each image. Statistical analysis was carried out using SPSS v18.0 and Sigmaplot v11.0.The baseline OAI radiographs were taken using 5 different sets of radiographic equipment and saved in DICOM format at varying resolutions. Pearson correlation was used to calculate relationships between the first mode from the appearance model (mode 1), continuous variables downloaded from the OAI database and ANOVA to test for significant differences between equipment manufacturers.
Results: The two cohorts were well matched using height and weight (P= 0.91 and P= 0.98 respectively [paired t test]); however the age matching was less successful due to the non-Gaussian distribution (P=0.28 [Wilcoxon Signed Rank test]).
There was a wide variation in the measured distance between knees between the manufacturers (P<0.00001), showing systematic differences between the radiographs. A new standardised measure (stDist) was found by subtracting the mean distance from each manufacturer.
Mode 1 was significantly correlated with Hip-knee angle (HKA), minimum medial joint space narrowing (MMJSN) on both sides (HKA P<0.001r=0.3; MMJSN P<0.00001 r=0.4 p>0.01). The distance between the knees was also significantly correlated with mode 1 (P<0.00001, R=-0.50). Despite this, no significant systematic difference was evident in mode 1.
The scatterplot of mode 1 against the distance between the knees highlights the systematic differences between the imaging equipment (Fig 1a). However this is ameliorated when the distance is standardised (Fig1b). The correlation between mode 1 and stDist (P<0.00001, r=-0.6) was stronger than that for the unstandardised distance. Investigation of apparent outliers in figure 1b identified individuals with strong asymmetry between the left and right knees in HKA and/or JSW.
Conclusions: This study shows that the AAM model of bilateral knee radiographs can model gross anatomical differences associated with OA, independently of imaging resolution. This result shows its potential as a robust model for investigating radiographic OA from a variety of image sources.
Original languageEnglish
Pages (from-to)S193-194
Number of pages2
JournalOsteoarthritis and Cartilage
Issue numberSupplement 1
Publication statusPublished - 1 Sept 2011


  • osteoarthritis
  • radiographs


Dive into the research topics of 'The Effect of Radiographic Positioning on Active Appearance Modelling of Bilateral Knee Osteoarthritis in the Osteoarthritis Initiative'. Together they form a unique fingerprint.

Cite this