TRANSFORM BASED FUSION ALGORITHM FOR MEDICAL IMAGES

Abstract

Medical imaging plays a very important role in several clinical applications including diagnosis, research, treatment and planning because of the recent development in the technology and instruments. To increase the accuracy of the clinical information for medical experts, different imaging modalities such as X-ray, computed tomography (CT), magnetic resonance imaging (MRl), positron emission tomography (PET) etc. are required to deal with the medical diagnosis and evaluations. Besides the wide range of imaging modalities, none of the imaging modality is able to reflect the complete clinical information (redundant as well as complementary). Therefore, combining of two or more different images obtained using different medical imaging modalities widely known as fusion has become a very promising and challenging research area. Using a fusion process, the different information contained in the different source images can be integrated into a single fused image. The fusion of multimodal medical images provides better comprehensive and clear information than any individual imaging modality. This is better to recognize the region of interest in the fused image rather than source images due to better localization and orientation of abnormalities. This dissertation work depicts the three different fusion approaches for two different types of medical images such as CT and MR image. All the fusion algorithms are implemented at the pixel level using different transformation techniques such as wavelet transform (WT), nonsubsampled contourlet transform (NSCT) and nonsubsampled shearlet transform (NSST). For implementing the proposed fusion approach, two registered CT and MR images are taken and then decomposed by the suitable transformation method. After decomposing the CT and MR images, separately, three different fusion rules are applied for low and high frequency subbands, individually. To investigate the performance of the fusion algorithms, nine different datasets of CT and MR images are utilized and the WT, NSCT based several fusion approaches are also used to compare the performances of the fusion approaches presented in this dissertation work. These approaches are implemented in the NSST domain with three different fusion rules such as (a) spatial frequency (SF) motivated PCNN (b) novel modified spatial frequency (NMSF) motivated PCNN, and (c) novel sum modified Laplacian (NSML) motivated PCNN. The results of the subjective analysis done by an expert show that the proposed fusion approaches provide better fused images. Besides the subjective analysis, the objective analysis is also done in terms of different performance measures such as mutual information (Ml), spatial frequency (SF), information entropy (EN), universal image quality index (Qo), edge quality index (QX), and standard deviation (STD). The comparative quantitative analysis of such aforementioned methods for different image datasets present the effectiveness of the NSST based fusion schemes in fusing the CT and MR images.