Improvement of Sievert Integration Model in brachytherapy via inverse problems and Artificial Neural Networks

Abstract

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.