Multimodal medical image fusion using fuzzy radial basis function neural networks

Most medical images are blurry for two reasons: one is that noise signal blurs the high frequency signal of image edge, the other one is the border of tumor with normal tissues cannot be very well defined on the images, therefore it is difficult for radiology experts to delineate image. It is necessary that an approach based on fuzzy inference fusion is used for coping with these difficulties. The paper puts forward a new method for multimodal medical image fusion, that fuzzy radial basis function neural networks (fuzzy-RBFNN) is used to perform auto-adaptive image fusion. Global genetic algorithm (GA) is employed to train the networks. In the experiments, samples sets include artificial blurred medical image. Experimental results show, comparing with other image fusion method, the proposed approach is more excellent for multimodal medical images both in visual effect of fused images and in objective evaluation criteria, especially for blurry source images.