抄録
Automatic segmentation methods of MR images have been developed for the cardiac surgery and the brain surgery. In these fields, Region Growing method has been used mainly. In this method, the core was inserted manually, and the pixel adjoining the core was judged whether it was homogeneous or not from its features based on image information. The core grew adding the homogeneous pixels, and the region of interest was obtained as the grown core.
It is available for orthopedic surgery and biomechanics to obtain the location and the orientation of bones and soft tissues in vivo. However, MR images including them could not be segmented by the former region growing method based on only image information. This is because those tissues had fuzzy boundaries on the image.
Thus, we used not only intensity and spatial gradient as image information but also location, size and complexity of the tissue to segment the MR images. The pixel adjoining the core was judged from three local features of the pixel; its intensity, gradient and location, and two global features of the core region; its size and complexity. Judgment was performed by Fuzzy Reasoning to allow their fuzzy boundaries. The homogeneous pixel was added into the core region. It grew into normal size and smooth shape under constraint of global anatomical features.
Using the present method, as an example, radius, ulna and interosseous membrane were segmented from the multi-sliced MR images of forearm. Segmented tissues agreed with the shape inserted manually by a medical doctor. As s result, three tissues containing different features on the MR image could be segmented by a single algorithm. It takes about 10sec per slice by using an engineering workstation.
