Premium
SU‐E‐T‐635: Improvement of Arrangement to Collimator Sizes of Helmet in Gamma Knife Using Artificial Neural Network
Author(s) -
Moghaddam L,
Setayeshi S
Publication year - 2011
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 - 10.1118/1.3612598
Subject(s) - collimator , isocenter , artificial neural network , monte carlo method , computer science , dosimetry , pixel , artificial intelligence , optics , nuclear medicine , mathematics , physics , statistics , medicine , imaging phantom
Purpose: To improve the dose distribution for Gamma‐Knife4C. The final collimators of beam in Gamma knife are located in Helmet and consist of 4 sizes in 4mm, 8mm, 14mm and 18mm. The physician uses one of these sizes individually depend on treatment plan and the location of isocenter in tumor. In this project the improvement of treatment plan suggested by using the four sizes Helmet collimator together simultaneously. We developed a procedure to arrange the collimators in helmet to get optimum dose in tumor. Methods: To achievement this goal, ANN (Artificial Neural Network) is used. The procedure consists of nonlinear program together with Artificial Neural Network (ANN) to predict the best design of collimators in helmet and the location of shots. Using different sizes of collimators change the ellipsoidal shape of shots to a non uniformed shape which can be flexible to the location of shots. ANN predicts the best arrangement of collimators in helmet. To design ANN, data for training, was obtained by Monte Carlo Simulation (MCNP 4C). Results: To test the Network, different dose of iso‐centre and also neighbor pixels were considered as input to the network and the arrangements of collimators in helmet were obtained from it. The answers were checked by MCNP 4C simulation by calculating the dose in the assumed area and comparing with the desired values which were considered as input to the network. We got the mean square error between 0.0008 and 0.01. Conclusions: By using this method, dose distribution in tumor has better coverage. The shape of each shot can be considered not only sphere but also other shapes to cover tumor especially in target boundaries. This capability could improve the treatment plan to reduce the number of shots along with optimum dose.