Genetic evolutionary taboo search for optimal marker placement in infrared patient setup

• Published in: Physics in medicine & biology Vol. 52; no. 19; pp. 5815 - 5830
• Main Authors: Riboldi, M, Baroni, G, Spadea, M F, Tagaste, B, Garibaldi, C, Cambria, R, Orecchia, R, Pedotti, A
• Format: Journal Article
• Language: English
• Published: England IOP Publishing 07.10.2007
• Subjects: Algorithms; Humans; Image Enhancement - methods; Image Interpretation, Computer-Assisted - methods; Infrared Rays; Male; Prostatic Neoplasms - diagnosis; Prostatic Neoplasms - radiotherapy; Radiotherapy, Computer-Assisted - methods; Radiotherapy, Conformal - methods; Reproducibility of Results; Sensitivity and Specificity
• ISSN: 0031-9155; 1361-6560
• DOI: 10.1088/0031-9155/52/19/006

In infrared patient setup adequate selection of the external fiducial configuration is required for compensating inner target displacements (target registration error, TRE). Genetic algorithms (GA) and taboo search (TS) were applied in a newly designed approach to optimal marker placement: the genetic evolutionary taboo search (GETS) algorithm. In the GETS paradigm, multiple solutions are simultaneously tested in a stochastic evolutionary scheme, where taboo-based decision making and adaptive memory guide the optimization process. The GETS algorithm was tested on a group of ten prostate patients, to be compared to standard optimization and to randomly selected configurations. The changes in the optimal marker configuration, when TRE is minimized for OARs, were specifically examined. Optimal GETS configurations ensured a 26.5% mean decrease in the TRE value, versus 19.4% for conventional quasi-Newton optimization. Common features in GETS marker configurations were highlighted in the dataset of ten patients, even when multiple runs of the stochastic algorithm were performed. Including OARs in TRE minimization did not considerably affect the spatial distribution of GETS marker configurations. In conclusion, the GETS algorithm proved to be highly effective in solving the optimal marker placement problem. Further work is needed to embed site-specific deformation models in the optimization process.