Towards automated brachytherapy film implant labeling using statistical pattern recognition

This paper presents a new approach towards automating the identification and labeling of brachytherapy sources. Brachytherapy is a cancer treatment method in which radioactive sources (seeds) are implanted into and around cancerous tissues. A common method for determining the radiation dose distribution delivered to the patient requires analysis of two orthogonal radiographs that show the interstitial or intercavitary seed placement. First, the seeds must be identified on the radiographs, and then the corresponding seeds must be identified on both films. Once this labeling process is complete the spatial coordinates of the seeds can be determined and the 3 dimensional dose distribution can be computed. For the type of implant studied, typically the number of seeds is between 50 and 250, and the seed distributions are dense. The methodology for automatic seed labeling consists of three steps. First, Multiscale Geometric Statistical Pattern Recognition (MGSPR) is used to identify treatment relevant objects in the radiographs. Second, a priori information must be processed to verify the integrity of the object identification by MGSPR and to resolve any object identification ambiguities. Third, the identified objects are projected into 3 dimensions, completing the labeling task, and allowing computation of the dose distribution