The role of Imperialist Competitive Algorithm in motion dataset optimization for ANFIS prediction model at external surrogates radiotherapy

Abstract

Introduction: At radiotherapy with external surrogates, a consistent prediction model is utilized to estimate tumor location as model output on the basis of external markers motion dataset as model input. In this study, Imperialist Competitive Algorithm is proposed to process and optimize external markers motion dataset for prediction model. The simplicity of prediction model based on the proposed algorithm results lesser targeting error with the least computational time. Materials and methods: a prediction model based on Adaptive Neuro-Fuzzy Inference System is utilized with database of 20 patients treated with Cyberknife Synchrony system. In order to assess the effect of proposed data optimization algorithm, two methods were considered. The prediction model is used with and without implementing Imperialist Competitive Algorithm. Then, targeting error of ANFIS model at these two methods is compared, quantitatively. Results: By implementing the proposed algorithm, the performance accuracy of ANFIS prediction model is remarkably improved by eliminating unnecessary and noisy inputs. Moreover, model simplicity factor could highly reduce model computational time this is required for clinical practice. Conclusion: Imperialist Competitive Algorithm was proposed as data optimization algorithm on motion dataset of patients who are treated with external surrogate’s radiotherapy. The proposed algorithm could highly optimize the external markers motion dataset as input of ANFIS model for estimating tumor position by selecting enough data points with high degree of importance. Final results show an improvement of targeting accuracy of prediction using proposed strategy and also significant reduction at model computational time.