A Novel Medical Image Fusion Scheme Using Weighted Sum of Multi-scale Fusion Results

Abstract

Fusion of medical images helps improve diagnosis and treatment by combining complementary data from different imaging modalities such as PET, MRI and CT. Several techniques for fusing medical images have been developed, but lack of contrast and distortion of fine details remain important concerns. In this paper, we propose a novel two step medical image fusion scheme. In the first step, individual multi-scale fusion techniques are applied to obtain fused images. In the second step, the individual results are combined using weighted average, with local structural similarity measure used as weights. In this way, a superior quality fused image is obtained. To evaluate the performance of the proposed scheme, several experiments were performed on PET-CT and MR-CT fusion. Experimental results show that the proposed scheme is capable of producing well-fused images as compared to individual multi-scale techniques i.e. discrete Wavelet transform, dual-tree complex Wavelet transform, Laplacian pyramid, Contour let transform and Curve let transform based fusion. Fused images were evaluated using multiple quality metrics. The proposed scheme demonstrated improvement of 3 to 4% in Mutual Information measure, around 2% in PSNR and 2 to 5% in a modified Universal Image Quality Index measure. Our results also scored well in other methods of evaluating fusion quality, namely Structural Similarity and Correlation.