|
The aim of this article is to optimize the structure of the artificial neural network (ANN) in order to obtain
the best imaging results. During the research, many variants of prediction models were trained, differing in the number
of neurons, the number of hidden layers, the learning algorithm, transfer functions, overfitting prevention method,
etc. As a result of comparing the results of the obtained reconstructive images with the reference images, the optimal
network structure was selected. Noteworthy is the original approach, which consists in training separate ANNs for
each voxel of an image separately. As a result, the model consists of many separately trained, single-output ANNs,
creating a complex system of models that constitute multiple neural network (MNN).
|
