Generalized regression neural networks for biomedical image interpolation

A neural-statistical approach to biomedical image interpolation using generalized regression neural networks is presented. These networks are basis function architectures that approximate any arbitrary function between input and output vectors directly from training samples, and with any desired degree of smoothness, and thus can be used for multidimensional interpolation. Experimental results compare favorably with other interpolation techniques. Because of their flexibility and ease of training, generalized regression networks can be used to complement existing approaches, and can be especially useful for post-registration image fusion and visualization.