PROCESSING OF MEDICAL IMAGES FROM LARGE DATASETS USING CONVOLUTIONAL NEURAL NETWORK

In this paper, a convolutional neural network trained on input-output pairs which show the operation of the operator. The original operator would not need to be run after testing. The qualified network functions completely and runs continuously. We study the impact on the approximation precision, runtime and memory footprint of network architecture and describe a certain architecture that combines these aspects. The same model applies to both operators. Experiments show that the method proposed is much more reliable than previous approximation schemes. We demonstrate that our models generalize across data sets and resolutions and explore a range of extensions of the method given.