Parameter estimation from realistic experiment scenario using artificial neural networks

One of the promising directions in future medicine is the optimization of therapies based on mathematical and engi-neering methods, with which the treatment can be personalized. In personalization, the key issue is the identification of the model parameters. We carry out the identification using artificial neural networks, which require training. We generate training data that is as realistic as possible, by generating sparse training datasets. We aim to simulate realistic experimental setups that also take the presence of holidays into account, when the drug injection and tumor volume measurement are not possible. We generate several smaller training datasets and their different versions, to determine how many times the parameters can change every week for the proper functioning of neural networks and to examine how the networks can give a proper estimate of the time-varying parameters from sparse measurements. In this research two different types of training datasets will be introduced to train artificial neural network. The training data are generated on known parameter intervals, taking into consideration the experimental setup we use to validate our results. The estimated parameters can be used to track the change of the parameters and personalize the model for optimization algorithms.