An algorithm for automatic, computed-tomography-based source localization after prostate implant

Permanent implant of the prostate using I-125 and Pd-103 seeds is a popular choice of treatment for early-stage prostate cancer in the United States. Evaluation of the quality of the implant is best based on the calculated dose distribution from postimplant computed tomography (CT) images. This task, however, has been time-consuming and inaccurate. We have developed an algorithm for automatic source localization from postimplant CT images. The only requirement of this algorithm is knowledge of the number of seeds present in the prostate, thus minimizing the need for human intervention. The algorithm processes volumetric CT data from the patient, and pixels of higher CT numbers are categorized into classes of definite and potential source pixels. A multithresholding technique is used to further determine the number of seeds and their precise locations in the CT volume data. A graphic user interface was developed to facilitate operator review of and intervention in the calculation and the results of the algorithm. This algorithm was tested on two phantoms containing nonradioactive seeds, one with 20 seeds in discrete locations and another with 100 seeds with small distances between seeds. The tests showed that the algorithm was able to identify the seed locations to within 1 mm of their physical locations for discrete seed locations. It was further able to separate seeds at close proximity to each other while maintaining an average seed localization error of less than 2 mm, with no operator intervention required.