Fusion Methods Of MRI & MRSI Based On Matrix Factorization

Nuclear magnetic resonance(NMR) is an important application of information technology in the field of medical diagnostics, winning the favor of the world with its non- invasive testing methods and accurate test results, which has become the most authoritative methods of detection in the medical field, especially in the detection of primary human brain tumors. Magnetic resonance imaging(MRI) and Magnetic resonance spectroscopic imaging(MRSI) are both vital applications of NMR, in which MRI has high spatial resolution but with low accuracy rate while MRSI has high accuracy but with low spatial resolution. Neither of the two types of data can satisfy the clinical requirement of resolution and accuracy. But a fusion result with high spatial resolution and high accuracy can be obtained by fusion method.The fusion of MRSI data and MRI data belong to multi- modal data fusion since MRSI data is spectral data and MRI data is the image data. Based on data fusion method, an unsupervised multi- modal data fusion method is proposed using non- negative matrix factorization and wavelet theory. The main contribution is presented as follows:1、In order to get a visual result of the MRSI data, a NMF method is used to decompose the MRSI data into a reference spectrum matrix and a coefficient matrix, which solves the problem that the MRSI can not provide an intuitive result. The spectra source data is transferred into a visual result by using NMF method.2、A multi- model fusion method based on wavelet theory is proposed to improve the disadvantage of the MRI and MRSI data. In each iteration of the NMF deco mposition, the intermediate results is fused with the MRI data and the fusion result is used to replace the original intermediate results of the decomposition. The fusion result with high space resolution and high accuracy is gained after the Matrix factorization.3、Because of that the acquisitio n and preprocessing of the MRSI data is difficult, the fusion method based on supervised fusion method is not appropriate to the fusion of MRI and MRSI data. Three different fusion rules are proposed and the in vivo experiment indicate that the fusion effect of adaptive weighting coefficients fusion method is ideal.