Samenvatting

Carpal tunnel syndrome (CTS) is the most commonly diagnosed peripheral neuropathy due to entrapment. This syndrome is currently being diagnosed clinically, often confirmed through the use of neural conductivity studies (NCS). Over 25 percent of the cases are still missed using this method, while NCS also cause discomfort to the patient. This is why other methods of diagnosis are being sought after, among which ultrasonographic imagery. It is thought the enlargement of the median nerve at the level of the carpal tunnel plays a role in the development of CTS. To further investigate this claim, the characteristics of the median nerve at this level are being investigated. A former study has been conducted to establish average values of these characteristics. For this study transversal cine loops of the proximal side of the carpal tunnel have been obtained from a population of 50 patients with CTS and 23 controls. Using software developed by the Erasmus Medical Center, the median nerve has been evaluated by manually placing tracing points around its perimeter. This method yielded the perimeter, cross-sectional area and circularity of the median nerve. Manual tracing however proves to be very time consuming and incorporates elevated levels of inter- and intra-rater variability. The assignment for this graduation internship was to develop an algorithm to automatically
determine the parameters mentioned above. Furthermore the reliability of this method had to be compared to the manual tracing method. The algorithm developed in this internship makes use of automated border detection, also known as minimal cost analysis or dynamic programming.
To facilitate the process of development and testing, a database has been developed to store the results in a hierarchy of patients and trials. The area, perimeter and circularity found by the algorithm were evaluated on its differences compared to the manual tracing method. Furthermore the amount of overlap between the areas found by the algorithm and the manual tracing method was assessed. The areas found by the algorithm approximately overlap the manual tracing data by 80%. Furthermore a trend is perceived where smaller shapes are overestimated, while larger contours are underestimated. Based on these findings, a number of improvements have been suggested. First of all the average size of the median nerve should be taken into account when trying to find the nerve's contour in order to make a better estimate. The algorithm is applied iteratively, as to use
the findings of earlier iterations as an approximation of the shape to be found. By adding an additional iteration, the algorithm's findings would come even closer to the actual shape. Finally,
by evaluating mean intensity gradients instead of local intensity gradients, noise is removed. This would improve the outcome of the algorithm.