Multi-Mode Medical Image Fusion Algorithm Based on Principal Component Analysis

Abstract

Based on research of principal component analysis, the principal component analysis is introduced to medical image fusion. The K-L transform is used to multi-mode images. Then a new matrix is composed. A eigenvector which accounts for above 90 percent in contribution of variance about the new matrix is adopted to obtain principal components. Using principal components can carry on image fusion. The result indicates that the method has many characters, such as fast execution, great information entropy and broad dynamic range.

t-SNE based Multi-mode Medical Image Fusion Algorithm

“…It can be clearly seen that the newly proposed method outperforms all the existing methods on all the three parameters. The second closest match is the fusion based on the principle component analysis [17]. Through comparing the value of average gradient, the average gradient of fusion image based on the ratio of t-SNE information and energy is much larger than that of the original CT image fig.…”

Advances in multimodal data fusion in neuroimaging: Overview, challenges, and novel orientation

“…The essence in statistics-based methods lies in the data-driven technique and high order statistics that can reveal the underlying pattern across multiple modes of data. Principal component analysis (PCA) [232] , [233] , [234] , [235] together with Hidden Markow Tree (HMT) [236] , [237] , [238] are two typical examples of statistic methods in the field of multi-modal medical image fusion.…”

Spiking Cortical Model Based Multimodal Medical Image Fusion by Combining Entropy Information with Weber Local Descriptor

“…Existing pixel-level image fusion methods generally include substitution methods, multi-resolution fusion methods and neural network based methods. The substitution methods such as intensity hue saturation [ 6 , 7 ], principal component analysis [ 8 ] based methods can be implemented with high efficiency but at the expense of reduced contrast and distortion of the spectral characteristics. Image fusion methods based on the multi-resolution decomposition techniques can preserve important image features better than substitution methods via the decomposition of images at a different scale to several components using pyramid (e.g., contrast pyramid [ 9 ] and gradient pyramid [ 10 ]), empirical mode decomposition [ 11 ] or various transforms including wavelet transform [ 12 , 13 , 14 ], curvelet transform [ 15 ], ripplet transform [ 16 ], contourlet transform [ 17 ], non-subsampled contourlet transform (NSCT) [ 18 , 19 , 20 , 21 , 22 ] and shift-invariant shearlet transform [ 23 , 24 ].…”