Cases

Osteoarthritis (OA) of the tibiofemoral (TF) joint is marked by distinct changes in cartilage volume and thickness. In vivo methods that accurately track these initial changes are crucial for documenting the onset and progression of OA. MRI technology shows promise as a tool for assessing those changes. It’s very important, however, to make use of accurate, reliable and efficient cartilage segmentation techniques so as to obtain trustworthy data. In order to find the best way to do this, the authors of this case assessed the reliability and accuracy of manual and semi-automated segmentation methods for quantifying knee cartilage thickness, both in and ex vivo. 3D laser scanning, considered to be the gold standard, was used as a benchmark method in ex vivo tests. The researchers wanted to compare only the most effective techniques in every method and therefore turned to Mimics to perform the manual segmentations. An adapted LiveWire algorithm was used as a semi-automated technique.

Manual segmentation as offered in Mimics allows the user full control over the segmentation process and the resulting 3D models. The end product, however, may differ depending on the user’s skills. Manual segmentation is still the preferred method for many researchers who want to control the segmentation process. Although Materialise invests constantly in improving Mimics’ segmentation speed and automating its segmentation process, while retaining user-friendliness, manual segmentation is still more time-consuming than (semi-)automated segmentation techniques, such as active contours or ‘snakes’. Although these techniques are usually faster, they are also prone to errors because they may miss localized features of cartilage damage and do not function well in regions of poor contrast. The researchers therefore opted for the LiveWire approach, which uses localized image intensity to directly map cartilage boundaries.

For this reason, it’s less likely to get trapped in local minima than other semi-automated techniques. The research team collected MR images of a healthy 23-year-old female for in vivo testing and identified specific regions of interest, which allows them to focus on primary loadbearing regions, where thickness changes are most significant, while excluding regions that are not affected and may mitigate these changes.

For the manual segmentation of the TF joint, the MRI images were imported into Mimics where 3D surface models were generated of each cartilage structure. Mimics’ highly efficient manual segmentation tools were central to the production of cartilage models that accurately represent the cartilage’s actual shape and volume.

The LiveWire algorithm was adapted to work in concert with manual segmentation so that the user could override the semi-automated method in regions where cartilageboundaries were unclear. The resulting contours were exported as point clouds that were used to generate closed-surface (solid) models for each structure. The solid models were constructed by wrapping the 3D point cloud with a triangular surface mesh.

The accuracy of both segmentation techniques was assessed on a cadaver specimen by comparing it to a 3D laser scan of the cartilage surface to obtain the true cartilage morphometry. This method was considered the gold standard for ex vivo comparison with the other techniques.

After thoroughly investigating both the reliability and accuracy of manual and semiautomated segmentation techniques, the research team concluded that both are repeatable and reliable methods in vivo as well as ex vivo. Laser scanning proved to be an acceptable gold standard for the evaluation of segmentation based cartilage thickness measurements. Manual segmentation conducted in Mimics was found to be more accurate than a third party's LiveWire approach, since it more closely approximated true cartilage thickness. The latter consistently underestimated the measurements attained with Mimics and laser scanning, making Mimics the most reliable and accurate virtual segmentation method.