Point-Based Registration of Brachytherapy Implants
Loading...
Date
2012-01-04
Authors
Gordon, Lauren Elizabeth
Keyword
Medical computing , Brachytherapy , Registration , Prostate cancer
Abstract
Prostate brachytherapy, a treatment for prostate cancer, was a procedure that typically involved placing radioactive sources in a cancerous prostate using percutaneous needles. The placement of these
sources determined the dose that the prostate and healthy tissues surrounding it received. However, because a needle could bend, tissue
could deform, and a patient could move, each source may have been displaced from its planned position. This source misplacement could later cause some cancer to be spared or healthy organs to be further
damaged. To better understand patterns of source misplacement, and eventually reduce the phenomenon, this work matched and registered implanted sources with their planned positions.
Each implant was registered to its plan using a sequence of four successive registrations. A rough initial registration was first found, using features known in the planned dataset and estimated from the implanted dataset. Second, subsets of sources were reconstructed
in the implanted dataset. The implanted sources were next matched to the planned sources using the subsets as constraints. Finally, the optimal rigid transformation between the implants and the plan was
found.
The algorithm was tested on both simulated and clinical datasets. Simulations placed limits on how properties of the subsets affected registration accuracy. When tested on 9 clinical datasets, the algorithm found 100% of correct plan-implant source matches within seconds on commonly available computers. When the implanted strands
were reconstructed as sine waves, 97% of t strands had an amplitude of less than 2mm. The clinical accuracy result generally agreed with simulation: subsets with amplitudes less than 2mm were expected to produce an accuracy >90%. The high accuracy of the algorithm may enable its use in finding patterns of source misplacement. The fast run-time of the algorithm may additionally make it useful for use in a clinical setting.