Neural network for maximum entropy restoration of nuclear medicine images
Huai D. Li, Weixian Qian, Laurence P. Clarke, Maria Kallergi · IEEE International Conference on Acoustics Speech and Signal Processing · 1993
A modified neural network which uses maximum entropy as a constrained condition is proposed to restore the degraded images obtained by the detection of bremsstrahlung radiation from beta emitters. The restoration consists of two steps: estimation of the parameter of the neural network and reconstruction of the image. In the first step, the neural network is modified so that the neuron states can represent the image gray level directly. In the second step, the energy function minimization procedure is modified to get the maximum approach. The restoration performance and stability of the method are evaluated by applying it to simulated images and clinical studies acquired with a gamma camera. The restoration obtained proved to be more stable compared with other approaches.>