Artigo de peri??dico
Improvement of Sievert Integration Model in brachytherapy via inverse problems and Artificial Neural Networks
Registro en:
0969-806X
155
SI
10.1016/j.radphyschem.2018.05.024
60.153
73.00
Autor
NASCIMENTO, ERIBERTO O. do
OLIVEIRA, LUCAS N. de
CALDAS, LINDA V.E.
Resumen
Increasing the radial distance, the accuracy of the Sievert Integration Model (SIM) decreases in a nonlinear
manner, adding errors up of 10% into the dose rate calculations?? a similar fact occurs to the 2D anisotropy
function where the errors may achieve 30% as already was related. For that reason, this paper sought an innovative
approach to optimize the error variance and its biases of dose rate calculations around a Nucletron
brachytherapy source of 192Ir from 0 to 10 cm taken in the radial distance, using an improved SIM through a
hybrid coupling of Artificial Neural Networks (ANNs) and Inverse Problem Theory (IPT). Since the traditional
approach relies into the use of a small data set of dose rate, the ANNs generalized these doses, making possible to
search more broadly optimum parameters to SIM using the IPT. The results showed excellent accuracy evaluated
with the Root Mean Square Percentage Error (RMSPE). In conclusion, the low RMSPE values indicate that the
methodology is consistent, showing an excellent agreement with the state of art of dosimetric measurement
techniques. Conselho Nacional de Desenvolvimento Cient??fico e Tecnol??gico (CNPq) CNPq: 165466/2015-4; 151013/2014-4; 301335/2016-8