Designing a Neural Observer to Estimate the State Variables of the Dynamical System of a Specific Class of Leukaemia

This article aims to present a novel neural network observer-based approach in order to estimate the state variables of the nonlinear dynamical system of chronic myelogenous leukemia (CML), specially the number of the infected cells. For this purpose, a two-layer feed forward neural network was applied. The weights of both layers are considered variables, depending on time. In order to adjust the neural network weights, the error back propagation learning algorithm was implemented. First of all, in this algorithm, the system outputs are generated according to random weights. Then the error is calculated and propagated back to the network and the weights are updated. This loop is executed until the error asymptotically converges to a small neighbourhood of zero. The better performance of a neural observer would be apparent in comparison with a classical high gain observer. Applying this method for estimating the state variables of cell dynamics results in a reduction in the number of tests and the required samples, which will consequently reduce costs and prevent wasting leukemic patients’ time.