Heat induction of iron oxide nanoparticles with rational artificial neural network design-based particle swarm optimization for magnetic cancer hyperthermia

This study presented the optimization of synthesis experimental conditions by examining the simultaneous effects of synthesis environment and precipitating agent type on the physiochemical, morphological, and magnetization characteristics of Fe3O4 nanoparticles (FeNPs). The experimental results implied that the effect of the synthesis environment outweighed that of the precipitating agent by enhancing the crystallinity and magnetization of FeNPs with smaller particle sizes. The hyperthermic performance of the optimal FeNPs was evaluated at different concentrations and different alternating magnetic fields (AMF). The promising candidate was 4 mg.mL(-1) of FeNPs under the AMF exposure of 15 kA.m(-1) and 320 kHz, according to the safe clinical criterion of Hxf (product) < 5x10(9) A.m(-1)s(-1), which produced desirable heat within the secure hyperthermia temperature range. Accordingly, the localized anticancer heat of optimal FeNPs was first examined using an artificial neural network (ANN) with three different ANN learning algorithms (namely BFGS Quasi-Newton (ANN-BFG), Levenberg-Marquardt, and Bayesian Regularization) to assess the highest performance accuracy. According to the results, the ANN-BFG model exhibited the best prediction performance with the smallest MSE and RMSE values. Thereafter, the particle swarm optimization (PSO) algorithm was successfully employed to amend the ANN-BFG speed rate. The hybrid ANN-PSO approach showed a remarkable potential mainly in predicting the effect of different influential parameters, including particle concentration, AMF product, and exposure time, on the localized temperature under hyperthermia conditions with excellent accuracy, convergence, and precise optimization according to the experimental or safe clinical hyperthermia setup.