Representative line detection algorithm with fuzzy inference and its application to segmentation of CT knee images

We propose a new algorithm called the representative line detection algorithm which embeds physician knowledge with fuzzy if-then rules. The algorithm detects the representative line of the region of interest (ROI). The representative line can show the rough location and shape. We first consider the representative line which consisted of some nodes. These nodes are then automatically detected by tracking the most suitable direction from the starting node. To evaluate this algorithm, it is applied to segmentation of the meniscus from CT knee images. The experimental results of six normal subjects showed that the representative line detection algorithm could express the center line of the meniscus, and could lead to detection of successful segmentation of the menisci.