Objective: Texture information of the subchondral bone area (SBA) of 2D radiographs represents a promising possibility for evaluating the state of osteoarthritis (OA). However such features are likely to vary within the SBA and therefore the selection of the region of interest (ROI) plays a crucial role. Thus, a feature selection algorithm (FSA) is being applied in order to determine ROIs that enable an optimum discrimination between patients with and without OA.
Methods: The study included 152 standardized knee radiographs from 66 cases and 86 controls. SBA was assessed by using both fractal analysis (Bone Structure Vale BSV) and a Shannon Entropy (SE) algorithm at predefined regions of the proximal tibia and the distal femur. The selected area of the proximal tibia involved a matrix of 3×8 ROIs, whereas a 2×2 matrix was defined for each condyle of the distal femur. SE and the BSV were calculated for each of the 32 ROIs, respectively. Based on these 64 variables, a FSA was applied to determine the variables that showed the best discrimination power.
Results: Combining the BSV and SE, the odds ratio increased significantly from 3.08 (95% CI: 1.785.30) to 14.82 (95% CI: 6.6932.83) when using 15 features, and to 39.75 (95% CI: 15.41102.51) based on ten features. By using the selected ten features the accuracy was found to be 0.86. This showed to be a significant improvement compared to the accuracy achieved when calculating a single mean value for the 3×8 ROIs of the proximal tibia alone (0.62 vs 0.86).
Conclusions: The application of a FSA in accordance with the combination of the two texture analysis methods shows a significant improvement with respect to the discrimination power between case and controls. The high odds ratios confirm that reliable results can be achieved by combining the BSV and the SE.
14 May 2016 - 17 May 2016