A novel medical image fusion algorithm for detail-preserving edge and feature extraction

By combining two or more medical images into one, image fusion has become an important tool for clinical diagnosis. However, existing fusion methods have also shown significant limitations, such as the loss of information content, weak contrast, noise and lengthy computation times. This paper presents a novel technique for medical image fusion that seeks to preserve and boost detailed information of the source images, while promoting its edges and textual features and suppressing noise. The method is based on a feature-linking, pulse-coupled neural network, followed by a modified Haar wavelet transform that leads to maximum-selection fusion in the transformed domain and high-scale Wiener filtering of the resulting image. The new algorithm is presented, described and evaluated on two sets of images, and its results are compared to those obtained from existing fusion methods. The performance of the newly developed algorithm is shown to be superior over the reference fusion methods in terms of a set of quality metrics based on subjective visual perception criteria, thus confirming its potential benefits to medical diagnosis.