Premium
Source localization from axial image sets by iterative relaxation of the nearest neighbor criterion
Author(s) -
Bice William S.,
Dubois Donald F.,
Prete James J.,
Prestidge Bradley R.
Publication year - 1999
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.598717
Subject(s) - imaging phantom , brachytherapy , projection (relational algebra) , algorithm , radiography , computer science , process (computing) , iterative reconstruction , k nearest neighbors algorithm , dosimetry , medical imaging , set (abstract data type) , computed radiography , computer vision , artificial intelligence , image (mathematics) , image quality , nuclear medicine , physics , optics , radiation therapy , medicine , radiology , programming language , nuclear physics , operating system
The use of axial image sets has become widely used to localize interstitial brachytherapy sources. One application of this method of localization is to perform post‐implant dosimetry following transperineal interstitial permanent prostate brachytherapy (TIPPB) where the target structure and the source locations are displayed on the same image. The design of an appropriate scanning sequence often results in abutting slices of an intermediate slice width (3, 4, or 5 mm). Because a single source may be imaged on more than one slice, the resultant scans always show many more source locations than actual sources implanted. The physicist is then faced with the tedious task of determining which sources appear on more than one slice and deciding which source locations to eliminate from the data set. We have developed an algorithm, similar to one employed by Roy et al., which automates this process by relaxing the nearest neighbor criterion until the number of sources is reduced to either the number of sources implanted or the number counted on a projection radiograph. This paper details this algorithm and the results of its application to phantom studies, comparing to known source locations, as well as clinical studies, comparing to orthogonal film source localization, on a series of ten patients. Phantom studies demonstrate the superiority of the algorithm over orthogonal film reconstruction, locating 100% of the sources within 5 mm of the actual location as compared to 66% for the paired radiographs. The clinical study findings are commensurate with these results, with 72% of the sources on average located within 5 mm of the corresponding source in the other data set. The positive correlation of the quality of the orthogonal film reconstruction results with the quality of the coregistration results suggests that differences in registration between the two data sets may be due primarily to the uncertainties in the orthogonal film reconstruction.