An automatic 3D detection method of seeds on CT images

Post-implant dosimetric evaluation is a key procedure of brachytherapy. The way of traditional manual localizing seed is lack of efficiency. Therefore, Automatic seed detection method is needed to efficiently and accurately calculate the centroid and orientation. There are some shape differences among seeds, caused by imaging features. Single seed may be presented on more than one slice and seeds locating closely may appear connected. This paper proposed an automatic three-dimensional detection method of seeds on CT images. Firstly, the areas possibly containing seeds are got by binary threshold on each CT slice and the related geometric information was recorded. Then, the larger areas containing more than one seed are segmented by watershed algorithm. According to the max seed volume rule and straight line rule, the areas are connected into the complete seed volume and the weighted centroid and orientation are calculated. The statistical analysis demonstrates that the rate of seed detection can achieve 97% and has a high applicability of post-implant dosimetric verification.