Fuzzy-logic adaptive neural networks for nuclear medicine image restorations

A novel neural network with adaptive fuzzy logic rule is proposed for image restoration as required for quantitative imaging using a nuclear gamma camera. The overall aims is to compensate for image degradation due to photon scattering and photon penetration through the collimated gamma camera to allow more accurate measurement of radiotracers in vivo. In this work, fuzzy rules are generated to train a membership function using a least mean squares (LMS) algorithm. This membership function allows one to describe rules to differentiate gray levels differences, so that the regularizer parameter can be optimally adjusted, based on the fuzzy membership value. The relative performance of this algorithm is compared to a previously reported neural network based hybrid filter by the authors (Wei Qian et al., 1993-8) using both simulated images with different noise levels and experimentally acquired nuclear images using a gamma camera. The improved signal to noise ratio (/spl Delta/SNR) is used for quantitative measurement. The proposed method has proved to be useful in quantitative imaging using a gamma camera for the planar and tomographic imaging mode using single photon emitters, beta emitters (bremsstrahlung detection) and positron 511 keV imaging.