Using artificial neural networks and model predictive control to optimize acoustically assisted Doxorubicin release from polymeric micelles.

We have been developing a drug delivery system that uses Pluronic P105 micelles to sequester a chemotherapeutic drug--namely, Doxorubicin (Dox)--until it reaches the cancer site. Ultrasound is then applied to release the drug directly to the tumor and in the process minimize the adverse side effects of chemotherapy on non-tumor tissues. Here, we present an artificial neural network (ANN) model that attempts to model the dynamic release of Dox from P105 micelles under different ultrasonic power intensities at two frequencies. The developed ANN model is then utilized to optimize the ultrasound application to achieve a target drug release at the tumor site via an ANN-based model predictive control. The parameters of the controller are then tuned to achieve good reference signal tracking. We were successful in designing and testing a controller capable of adjusting the ultrasound frequency, intensity, and pulse length to sustain constant Dox release.