Prostate brachytherapy seed localization by analysis of multiple projections: identifying and addressing the seed overlap problem.

Intraoperative three-dimensional reconstruction of seed locations during prostate brachytherapy for purposes of immediate computation of radiation dosimetry is an active area of current investigation, including methods which use multiple fluoroscopic projections. A simulation study using seed locations extracted from clinical CT data was performed; the result showed that on average one quarter of the seeds had a projection image overlapping with other seeds. The average percentage of non-overlapping seeds for the prostate implants and seed types investigated was 74.5% with a range of 56.9%-92.9%. The distribution of seeds in different cluster sizes was analyzed as well as the distribution of pixel counts of connected components. A statistical classifier was developed to determine the number of seed images in a self-connected component in the segmented images. The classifier was tested with simulation data, and the error rate was below 2%. A method to determine seed image position is also provided. A modified three-film technique was used to reconstruct 3-D seed locations. The algorithm allows unequal number of seed images for each projection as input while current methods require the same number of seed images detected in all projections. An accuracy analysis based on angular and positional uncertainty was performed. The reconstruction and seed localization algorithms were tested with simulation data, and the mean distance error of the reconstructed results was 0.61 mm. A phantom study was performed to validate the seed localization method. Three false positive seeds, 4.7% of the total, in the reconstruction result were observed in this study.