Neural network design
Adaptive control using neural networks and approximate models
IEEE Transactions on Neural Networks
| Hi-index | 0.00 |
This paper describes a software system for hemodynamic simulation of the blood flow through an aneurysm with rigid walls. Human health significantly depends on the status of blood vessels, however measuring in vivo the important hemodynamic factors in the arterial blood flow, such as wall shear stress, pressure and velocity in blood vessels, is very difficult. Finite Element Mesh (FEM) of the observed blood vessel is a starting point in a patient specific blood flow simulation and measurement of these hemodynamics. A feed-forward neural network is proposed here that performes required computations in each point of the predetermined geometry of the mesh. The architecture of that neural network is two layers with sufficient number of neurons in the hidden layer. Optimal value is empiricaly determined. The first step in the training process is assembling input vectors and the corresponding target vectors, obtained as an output of MedCFD application. Among many variations of backpropagation training functions implemented by generalizing the Widrow-Hoff learning rule, created network object uses Levenberg-Marquardt algorithm. Training continues until the network can associate input vectors with appropriate output vectors. Finally, network is able to simulate and give good responses to new inputs.