MULTIMODAL MEDICAL IMAGE FUSION USING DUAL TREE- CWT AND NON-SUBSAMPLED CONTOURLET TRANSFORM 1 K.Kandasamy, V.Manikandan, 3 M. Rajaram, 4 K.S.Tamilselvan 1 Anna University-Villupuram campus, India. Coimbatore Institute of Technology, Anna University, Chennai, India. 3 Anna University, Chennai, India. 4 Velalar College of Engineering and Technology, Anna University, Chennai, India. ABSTRACT The goal of Multimodal Fusion techniques is to integrate complementary information from different sensors together to produce a more accurate and efficient representation in a single fused image which is more suitable for the purpose of human visual perception and for further analysis. Fusion is a specific method within many vital fields such as remote sensing, robotics and medical applications. In this paper, the multimodal images are decomposed using the Dual Tree Complex Wavelet Transform (DT-CWT) and Non-Subsampled Contourlet Transform (NSCT) is applied to the decomposed images and they are fused using efficient and robust fusion rules. Finally, they are reconstructed using the Inverse transform and a new fused image is obtained with more information content. Experiments showed that the proposed fusion technique can have better performance of fusion is quantitatively measured using Peak signal to Noise Ratio (PSNR), Entropy, and Mutual Information. Keywords- Fusion, Discrete Wavelet Transform (DWT), Dual Tree Complex Wavelet Transform (DTCWT), Non-Subsampled Contourlet Transform (NSCT). INTRODUCTION The aim of image fusion is to integrate matching information from different images to create a highly informative image which is more suitable for human visual perception or computer-processing tasks. Some applications of the fusion of images include medical imaging, microscopic imaging, remote sensing, computer vision, military applications and robotics etc. Fused images may be created from multiple images of the same imaging modality, or by combining information from multiple modalities, such as Computed Tomography (CT), Magnetic Resonance (MRI), Positron Emission Tomography (PET), and Single Photon Emission Computed Tomography (SPECT). Functional image such as computed tomography (CT) can provide information about dense structures like bones and implants with less distortion, but it cannot detect physiological changes, whereas PET has low spatial resolution and provides information about blood flow activity. In anatomical image such as MRI image soft tissues can be better visualized. Medical fusion is to combine functional image and anatomical image together into one fused image which provides abundance information to doctor to diagnose clinical disease and also reduces the storage cost by reducing storage to a single fused image instead of multiple-source images. In recent years, a lot of researchers have paid their attentions on the field of multi-resolution region-based image fusion. Overview of Fusion The medical image fusion mainly uses pixel based fusion techniques. Usually, the pixel level fusion is broadly classified into three main categories, they are as follows, 1. Spatial Domain Techniques (PCA, Averaging, etc.): The main advantage of this technique is, they are easy to implement. But has a few drawbacks that it produces spatial distortion in images. Also, in the final fused image some of the image details will not be present with respect to the input images.. Optimization Approach (Bayesian Approach): It suffers from the problem of computational complexity. 3. Transform Domain Approach (Multi-resolution Techniques): This transform preserves the structural characteristics and also it provides more information for further analysis and diagnosis of various diseases. The successful fusion process should extract complete information from the source images and preserve all that (relevant) information in the resultant fused image without introducing any artifacts or inconsistencies. Earlier, multiscale decomposition using wavelet transform has been widely used and identified as ideal method for image fusion because it minimizes structural distortions and good at isolated discontinuities, but it suffers from shift sensitive, poor directionality and absence of phase information. To overcome these drawbacks Dual Tree Complex Wavelet Transform (DTCWT) and Nonsubsampled Contourlet Transform (NSCT) is proposed. DTCWT based multimodal medical image fusion has approximate shift invariance, good directional selectivity and perfect reconstruction with limited redundancy. The NSCT is shift-invariant, multiscale, and multidirectional transformation where Contourlet represents the long edges very well and it extracts the geometric information of images very well and fast implementation which is useful to many image processing tasks. Discrete wavelet Transforms (DWT) The Discrete Wavelet Transform (DWT) is spatial frequency decomposition that useful for multi resolution analysis of an image. -D Discrete Wavelet Transform uses a separate filter and down sampling in the horizontal JCHPS Special Issue 9: April 015 www.jchps.com Page 9
and vertical directions produces four subbands in each scale. Denoting the horizontal frequency and then the vertical frequency which produces high-high (HH), high-low (HL), low-high (LH) and low-low (LL) image subbands. The left-top sub-image is the band with low rang a frequency that has the lowest spatial resolution and point out the approximation information of the original image. Whereas the other sub-images (the bands with high frequencies) show the detailed information of the original image. At each scale the subbands LH, HL and HH subbands are sensitive to vertical, horizontal and diagonal frequencies respectively. Multiresolutional decomposition can be achieved by recursively applying the same scheme to low-low subband. Wavelet based image fusion is the multi scale (multi resolution) approach well suited to manage the different image resolutions. DWT allows the decomposition of image in to different coefficients and such coefficients coming from different images are combined appropriately in to single fused image and taking the inverse discrete wavelets transform (IDWT) for this image, the final fused image is obtained, where the information of the individual input images is preserved. Fig 1 Wavelet based image fusion Although DWT based techniques provide better results for image fusion than pyramid transform and other spatial domain method, yet it suffers from some of drawbacks are shift sensitive and a low directional transform, absence of phase information and uses very redundant representation. To overcome these drawbacks we are going for proposed methods Dual Tree Complex Wavelet Transform (DTCWT) and Non Subsampled Contourlet Transform (NSCT). Dual-Tree Complex Wavelet Transforms (DTCWT) DTCWT beats the downsides of DWT. The Dual tree Complex wavelet Transform (DT-CWT) is complex esteemed expansion of the standard wavelet. Complex Wavelet Change utilizes complex esteemed separating that disintegrates the info image into true and missing parts in change area in which their comparing coefficients give greatness and stage data. DT-CWT produces shift invariance, which can accomplish in DWT by multiplying the inspecting rate. This is effected in the DT-CWT by wiping out the down testing by after first level channel. Two completely wrecked trees are then created by down inspecting, effected by taking first even and after that odd samples after the first level of filters. To get uniform interims between the two tree's samples, the resulting channels require a large portion of a specimen distinctive postpone in one tree. Application to picture can be attained to by detachable complex sifting in two measurements. The real -D dual-tree DWT of a image x is executed utilizing two basically inspected distinguishable - DDWTs in parallel. At that point for every pair of subbands we take the whole and distinction. The complex -D DT-DWT likewise offers ascend to wavelets in six particular bearings. The complex -D double tree is actualized as four discriminatingly examined detachable -D DWTs working in parallel as demonstrated in figure 3. -D structure needs four trees for investigation and for combination. The sets of conjugate channels connected to two dimensional images (x, y) can be communicated as: (hx+jgx ) (hy+jgy )= (hx hy - gx gy) +j (hx gy +gx hy) (1) DT-CWT improves the directional selectivity than DWT, because DTCWT gives rise to wavelets in six distinct directions, ±15, ±45, ±75 and whereas DWT have three subbands in 0, 45 and 90 directions only. JCHPS Special Issue 9: April 015 www.jchps.com Page 10
Fig Complex Wavelet Transform Scale Fig 3 Filter bank structure for -D DT-DWT Orientation labeled subbands Non-Subsampled Contourlet Transform (NSCT) The NSCT is a shift invariant version of the Contourlet transform. NSCT is the combination of Non- Subsampled Pyramids (NSP) for multi-scale decomposition and Non-Subsampled Directional Filter Banks (NSDFB) for directional decomposition. The NSP is a two-channel non-subsampled channel bank, and one lowrecurrence image and one high-recurrence image can be created at every NSP decay level. The multiscale decomposition can be achieved by iterating using the Non-Subsampled filter banks. Such expansion results in k+1 sub-images, which consists of one low- and high-frequency images having the similar size as the resource image where k denotes the number of decomposition levels.nsp decomposition with k=3 levels is as shown in fig 4 The corresponding filters of a k th level cascading NSP are given by, n n 1 j I I H1z H 0 z 1 n k, j0 () eq H n z n j I 0 H z n k 1. j0 The NSDFB is two-channel non-subsampled filter banks which are constructed by combining the directional fan filter banks.nsdfb allows the directional decomposition with l stages in high-frequency images from NSP at each scale and produces l directional sub-images which provides directional information. Figure 5 illustrates four channels NSDFB construct with two-channel fan filter banks. eq D U z U z U z (3) The corresponding filter in each channel is given by k i j Fig 4 Three-stage non-subsampled pyramid decomposition. Fig 5 Four-channel non-subsampled directional filter bank. Fig 6 Decomposition frame work JCHPS Special Issue 9: April 015 www.jchps.com Page 11
Proposed fusion algorithm 1. Read the source images A (CT) and B (MRI).. Find the Dual Tree Complex Wavelet Transform (DTCWT) for the source images individually. 3. Apply Non Subsampled Contourlet Transform (NSCT) for the above decomposed images individually. 4. Apply Fusion rules, the low frequency coefficients follow the phase congruency and high frequency coefficients follow the largest absolute value rule. 5. Take the inverse transform for the fused image, INSCT followed by IDTCWT in order to obtain the final proposed fused image. The quality of the fused image is determined using the following performance measures. Fig 7 Flowchart of the FRFT-NSCT image fusion Performance Measure for Segmentation Efficiency i) Information Entropy (IE): The IE of the image is an important index that refers to the richness of image information quantity. The evaluation IE of the image is based on the Shannon information theory and it is given by, E m p i log p i i1 (4) Where p i is the ratio of the number of pixels with gray value equal to i over the total number of the pixels. The larger the value of IE reflects the more information carried by images and hence the fusion performance is increased. ii) Peak Signal to Noise Ratio (PSNR): PSNR is the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation. The PSNR is used to measure the quality of reconstructed images and it is given by:- PSNR( db) 0log M 55 3MN N B' i 1 j 1 i, j Bi, j (5) Where, B - the perfect image, B - the fused image to be assessed, i pixel row index, j Pixel column index, M, N- No. of row and column. iii) Mutual Information (MI): Mutual information (MI) is a quantity that measures the mutual dependence of two random variables. It usually shows measurement of the information shared by two images. Here, the information that reference and the fused image shares: L L P F i, j RI MI F P i j RI F, log (6) RI P i P F j i1 j1 R I Where P RI is the normalized joint gray level histogram of images R and I f F, P R and P I are the normalized marginal histograms of the two images. The mutual information IAF between the sources images A and the fused image F is defined as follows: PAF a, f I AF PAF a, f log P a P f AF A F (7) Where P AF is the jointly normalized histogram of A and F, P A and P F are the normalized histogram of A and F, and a,f represent the pixel value of the image A and F, respectively. The mutual information I BF between the source JCHPS Special Issue 9: April 015 www.jchps.com Page 1 MI F RI measures
image B and the fused image F are similar to I AF. The mutual information between the source images A, B and the fused image F is the sum of I AF and I BF, i.e. MI Sl. No. AB F I AF I BF (8) Table.1. Comparison of PSNR, Entropy and Mutual Information for Wavelet based Fusion Techniques and Non Subsampled Contourlet Transform Feature Inverse Dual Tree Inverse Non-Subsampled Set Complex Wavelet Contourlet Transform( Proposed Method Transform NSCT) 1 Entropy Set1 7.315 7.4896 7.516 Set 5.9957 6.8305 6.9517 Set3 6.3538 7.073 7.3607 Set4 6.0574 7.5338 7.6395 PSNR Set1 0.40 1.09.06 Set 1.98 3.10 4.07 Set3 3.57 5.4 6.68 Set4 4.06 4.79 5.61 3 Mutual Information Set1.047.136.063 Set.903 3.364 3.3964 Set3.063.197.958 Set4.016.6410.7810 Sl. No. Set Fused image using Inverse Dual Tree Complex Wavelet Fused image using Inverse NSCT Proposed image fusion Set-1 Set- Set-3 Set-4 CONCLUSION This paper presents a robust and efficient approach for the segmentation of noisy medical images. The proposed approach makes use of Inverse Dual Tree Complex Wavelet Transform (IDT-CWT) and Inverse Non- Subsampled Contourlet Transform (INSCT) image fusion can be performed to achieve a better image fusion quality compared with other techniques, especially for medical images. The experiments with original clinical JCHPS Special Issue 9: April 015 www.jchps.com Page 13
Brain CT images have been demonstrated. The efficiency of the proposed approach in segmenting the noisy medical CT image is increased to 96.5% which is greater than the existing techniques. With the knowledge of all types of existing image fusion techniques, they decide to continue the research work in the future with the application of Multi wavelets in combination with Fuzzy and Neural Networks. It also planned to develop a hardware implementation system suitable for the above fusion technique using FPGA. REFERENCES Abdullah-Al-Wadud.m, Md. Hasanul Kabir, M. Ali Akber Dewan, and Oksam Chae, A dynamic histogram equalization for image contrast enhancement, IEEE Transactions. Consumer Electron., 53(), 0007, 593-600 Balster. Eric I. nd Yuan E Zheng, Fast, Feature-Based Wavelet Shrinkage Algorithm for Denoising, KIMAS 003. October 1-3,003,Boston.,USA,Copyright 0-7803-7958-6/03/ 003 IEEE. Ashok Saini, Reduction of Noise from Enhanced Using Wavelets, International Journal of Electronics Engineering, 3 (), 011, 75 77. Candès.E.J and L. Demanet, Curvelets and Fourier integral operators, C. R. Math. Acad. Sci. Paris, 336 (003), 395 398. F. E. Ali, I. M. El-Dokany, A. A. Saad, and F. E. Abd El-Samie, Curvelet fusion of MR and CT images, Progr. Electromagn. Res. C, 3, 008, 15 4. Fowler.J, The redundant discrete wavelet transform and additive noise, IEEE Signal Processing Letters, 1(9), 005, 69 63. L. Yang, B. L. Guo, and W. Ni, Multimodality medical image fusion based on multiscale geometric analysis of contourlet transform, Neurocomputing, 7, 008, 03 11. N. Boussion, M. Hatt, F. Lamare, C. C. L. Rest, and D. Visvikis, Contrast enhancement in emission tomography by way of synergistic PET/CT image combination, Comput. Meth. Programs Biomed., 90(3), 008, 191 01. Q. Miao, C. Shi, P. Xu, M. Yang, and Y. Shi, A novel algorithm of image fusion using shearlets, Opt. Commun, 84(6), 011, 1540 1547. Q.Guihong, Z. Dali, andy. Pingfan, Medical image fusion by wavelet transform modulus maxima, Opt. Express, vol. 9, pp. 184 190, 001. R. Redondo, F. Sroubek, S. Fischer, and G. Cristobal, Multifocus image fusion using the log-gabor transform and a multisize windows technique, Inf. Fusion, 10(), 009, 163 171. S. Daneshvar and H. Ghassemian, MRI and PET image fusion by combining IHS and retina-inspired models, Inf. Fusion, 11(), 010, 114 13 S. Das, M. Chowdhury, and M. K. Kundu, Medical image fusion based on ripplet transform type-i, Progr. Electromagn. Res. B, 30, 011, 355 370. S. Li, B. Yang, and J. Hu, Performance comparison of different multiresolution transforms for image fusion, Inf. Fusion, 1(), 011, 74 84. S. Yang, M. Wang, Y. Lu, W. Qi, and L. Jiao, Fusion of multipara metric SAR images based on SWnonsubsampled contourlet and PCNN, Signal Process., 89(1), 009, 596 608. T. Li and Y.Wang, Biological image fusion using a NSCT based variable-weight method, Inf. Fusion, 1(), 011, 85 9. V. Barra and J. Y. Boire, A general framework for the fusion of anatomical and functional medical images, Neuro, 13(3), 001, 410 44. Y. Chai, H. Li, and X. Zhang, Multifocus image fusion based on features contrast of multiscale products in nonsubsampled contourlet transform domain, Optik Int. J. Light Electron Opt., 13(7),01, 569 581. Y. Yang, D. S. Park, S. Huang, and N. Rao, Medical image fusion via an effective wavelet-based approach, EURASIP J. Adv. Signal Process, 010, 44-1 44-13. JCHPS Special Issue 9: April 015 www.jchps.com Page 14