Using the measurement data, the generation of the 2D WearMap for a dataset of 56 segmented images typically took less than 5 s, on a standard personal computer. With the cartilage data stored in an array, three types of plots – the thickness graph, the 2D WearMap and the 3D WearMap – were generated for each dataset. The cartilages segmented from the images were measured and were visualized using the visualization framework described in Methods section. This is because each slice in the OA dataset had to be more carefully scrutinized to identify the actual cartilage within the image due to a decrease in cartilage contrast and homogeneity of the articular cartilage with OA. Segmenting OA data required more observer intervention in the semi-automatic segmentation process. Segmenting each dataset (i.e., a femur of one knee in one patient) from patient with normal knee generally took 10–15 min, whereas segmentation of the OA datasets generally took 30–40 min. MR images of all the patients were segmented using the semi-automatic segmentation method to delineate the femoral articular cartilage. Furthermore, the WearMap can be exported and stored in JPEG format for future reference. This allows the TrackBack of multiple slices. This function could be performed multiple times for different locations of the WearMap. The graph is displayed beside the MR slice and the calculated thickness value at the point selected on the 2D WearMap is shown on the thickness graph. This is achieved by creating a thickness profile for the specific slice to which the selected point belongs and by plotting thickness values against angular position. Additionally, a thickness graph can also be shown. (This is a function of the way in which the cartilage boundaries are sampled on segmentation.) To determine the thickness for any intermediate point on the 2D WearMap the thickness values based on 4° increments are interpolated, by linear interpolation, to the equivalent of 1° increments. Thickness values are available at 4° increments along each cartilage slice. Once the original MR image is identified, the thickness value at the point selected on the 2D WearMap is calculated. This approach is designed to cause the user to be drawn to regions where possible cartilage degeneration is occurring. The use of a log scale allows more colors to be allocated to thicknesses below 2 mm – to highlight smaller differences in cartilage thickness (0.25 mm rather than 1 mm). In contrast, thicknesses smaller than 2 mm are represented by a spectrum of red–black shades to highlight regions of cartilage wear or possible lesions. It is important to note that thicknesses in this color range are still within the expected normal thickness values. This color is chosen because the thicknesses in this range are approaching the thickness values (i.e., smaller than 2 mm) that represent cartilage wear. As the thicknesses get smaller (i.e., between 2 and 3 mm) a shade of yellow is used. Referring to Fig. 5, for thicknesses above 3 mm a shade of green is employed to indicate that the cartilage is within the thickness range that represents no concern. This allows the user to easily distinguish between normal and worn cartilage. This validation method is comparable to the work by Duryea et al.Ī color scheme was developed to represent cartilage thickness in the 2D and 3D WearMaps (discussed below). This further enhances the validity of the reproducibility of our segmentation method. Representative COV values for cartilage volume variation in the segmented datasets for each respective knee were also derived. The volume of the femoral articular cartilage for each segmented dataset was also calculated. We derived the standard deviation (SD) and the coefficient of variation – COV – which is defined as 100×(SD/average) of cartilage thickness from the mean cartilage thickness values calculated from each respective knee. In addition, the mean cartilage thickness values for each of the segmented datasets for the same knee were calculated. This method of validation allows the MR images to be compared pixel by pixel. The resulting values from each comparison of the datasets (for each respective knee) were averaged to derive the representative mean values of sensitivity and specificity, in percentage terms. Within the four semi-automatically segmented datasets for each patient, in order to establish reproducibility we directly compared the individual corresponding segmented images in each dataset to derive sensitivity and specificity values.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |