Pixel-level Medical Image Fusion

With the development of modern medicine, many medical image devices have been conceived. At the same time, different types of medical images are enhancing their applications in various fields. In the field of medicine, for instances, it's helping to make more accurate and comprehensive diagnosis and treatment. The comprehensive and reciprocal information could be obtained from different medical images within one pathological organization. The medical image fusion technology, as a branch of data fusion, could utilize the redundant information to improve the signal-to-noise ratio so that giving a more reliable result as well as to fuse the reciprocal information in order to obtain a composite image with more detail and complete content. Among fusion methods of different kinds, the Pixel-level Fusion method, which can best keep information from original images, is the prevalent method among professionals and the major study of this dissertation. The major contents of this dissertation are as follow.Firstly, the significance and background of the medical image fusion are proposed in this dissertation. In addition, on the bases of the brief introduction to the commonly used medical image, the development background of image fusion technology, the concept of image fusion and its research value in medical domain are indicated as well. In order to estimate the fusion result, a combination of the subjective and objective evaluation criterion for the performance of image fusion and the effect of the fused images, which is proved reasonable, is put forth.Secondly, some current image fusion algorithms in spatial domain such as: max-gray and min-gray fusion algorithms, constant weighting image fusion algorithm, partial variance weighting fusion algorithm, partial energy weighting fusion algorithm and contrast modulation fusion algorithm are introduced in this thesis. Based on the current contrast modulation fusion algorithm, two improvements are suggested in this dissertation: (1) choosing the image, which contains more detail information, to be modulated, (2) using the reversed-gray images, which could be obtained by converting the gray-value of the original images, to practice fusion. Through experiments, fusion results and the evaluation of subjective and objective are given to show the better performance of the improved contrast modulation fusion algorithm among that of the above algorithms.Thirdly, With regard to the method in wavelet domain, after introducing wavelet transformation elementary theory and the general wavelet fusion principle, the max-coefficient and min-coefficient fusion algorithms, the wavelet coefficient constant weighting fusion algorithm and the partial coefficient variance weighting fusion algorithm are studied in detail. Through the experiment, this dissertation point out the disadvantage of the traditional algorithms that the importance of the fusion algorithm in low frequency coefficients and the information of neighbor region in high frequency coefficients are not sufficiently taken into account, thus failing to get perfect fusion effect in some circumstances. To conquer this weak point, a new wavelet image fusion algorithm (Low and High frequency bands Partial Energy fusion algorithm: LHPE) is proposed. By employing different masks, the algorithm integrates the low frequency band coefficients and the high frequency band coefficients differently according to the weights, which are the energy of thecorresponding neighbor region of the wavelet coefficients. Reconstructed from the integrated coefficients, the fusion result could be obtained with more information of the original images. Evaluations of the experiment results according to both the subjective and objective criteria demonstrate that the proposed algorithm is more effective than the traditional algorithms.