Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Image Fusion of MRI and CT Images

Using DTCWT
Neelabh Saxena1 Pranav Jain2
Department of Computational Intelligence, Department of Computational Intelligence,
SRM Institute of Science and Technology SRM Institute of Science and Technology
Chennai, India Chennai, India

Dr. N. Meenakshi3
Assistant Professor
Department of Computational Intelligence,
SRM Institute of Science and Technology,
Kattankulathur Campus

Abstract:- Fusing different medical Images will increase To overcome these limitations, the suggested technique
the accuracy in diagnosis of disease and describe the employs Dual Tree Complex Wavelet Transform (DTCWT),
complicated relationship between them for medical which offers better directionality and shift variance, to
research. The current procedures take a lot of effort and combine various medical Images, making it easier to process
also requires additional data points for the models to be edges and contours of the source Image. The increased shift
trained. In this model, Through the use of multi-stage variance and directionality features of the DTCWT make it a
fusion networks, we will combine various medical fulfilled Image fusion tool, playing a significant role in
Images such as Magnetic Resonance Imaging (MRI) and theoretical and analytical tools for Image and signal
Computed Tomography (CT) to recover the intricate processing.
information. In the proposed model, we will use Dual
Tree Complex Wavelet Transform (DTCWT) to extract Overall, medical Image fusion helps doctors make
the complicated and correlated information from each better and more accurate treatment decisions by providing
Image and segmentation is done on fused Image to get more comprehensive and accurate information for diagnosis.
the segmented Image. In the proposed method, the
improved quality of final fused Image can be obtained. II. LITERATURE REVIEW

Keywords: Image Fusion, Dual Tree Complex Wavelet,  P. Kanimozhi, N. A and N. Padmapriya, "Improved
MRI, CT Scan, Medical Images, Segmentation. computer vision analysis for MRI and CT Images using
multistage fusion networks," 2022 International
I. INTRODUCTION Conference on Smart Technologies and Systems for Next
Generation Computing (ICSTSN), 2022:
Systems for medical imaging like X-rays, PET, SPECT, In this paper, they proposed fusing medical pictures
MRIs, and CT, provide valuable clinical data, but they often improves illness diagnosis and medical research by
fail to reveal detailed tissue information or provide a describing their complex interaction. Existing methods take
comprehensive diagnosis. To extract complicated a long time and require a lot of data. The proposed method
information from a single Image, medical Image fusion would use multi-stage fusion networks to extract complex
combines different types or similar types of medical Images information from MRI and CT Images. The proposed
to create a more accurate Image for diagnosis, which helps method first segments the fused Image using Dual Tree
doctors extract detailed and complicated data not visible in Complex Wavelet Transform (DTCWT) to remove complex
individual Images. Medical Image fusion is used in and related information from each Image. The suggested
oncology and cancer research therapy to create a perfect strategy improves final fused picture quality.
Image fusion process that gathers complementary
information from source Images while neglecting unwanted  R. Zhu, X. Li, X. Zhang and J. Wang, "HID: The Hybrid
and unexpected features. Image Decomposition Model for MRI and CT Fusion,"
in IEEE Journal of Biomedical and Health
The source picture needs to be pre-processed, Informatics,2022:
registered, and aligned before Image fusion. Three layers The NSST inverts the picture when shearlet transform
can be used for Image fusion: pixel, characteristic, and divides anatomical pictures (NSST). Low frequency is split
decision. The most common technique for Image fusion is using structural similarity and structure tensor optimization.
Discrete Wavelet Transform (DWT), which is flawed, having HID uses base-layer texture details to create fused low
shift variation and bad directionality, leading to errors in frequency Images. 12 innovative methods are contrasted
fused Images and difficulties processing geometric features. with the proposed strategy using 50 MRI and CT Images.

IJISRT23MAR1668 www.ijisrt.com 2493

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Fusion rules use base-layer texture details. Models extract lacks distortion. This suggested technique combines MRI
texture better. and CT brain imaging. This method's fusion outcomes are
quantified. This approach's hybrid fusion reaction shows the
 B. S. N. Rao, N. V. K. Raju, M. Dhanush, P. N. S. M. findings' supremacy.
Harshith and M. J. Mehdi, "MRI and Spect Medical
Image Fusion using Wavelet Transform," 2022 7th  J. Du, W. Li and H. Tan, "Three-Layer Image
International Conference on Communication and Representation by an Enhanced Illumination-Based
Electronics Systems (ICCES), 2022: Image Fusion Method," in IEEE Journal of Biomedical
By combining pictures from different imaging devices, and Health Informatics, 2020:
such as computed tomography (CT), magnetic resonance This study proposes a three-layer picture
imaging (MRI), and PET, medical picture fusion aims to decomposition method that uses rules for CT and MRI
increase Image clarity. This paper provides a 2-Dimensional fusion. Three phases are suggested. Each input picture is
DWT-based discrete wavelet transformation Image fusion deconstructed using local extrema and low-pass filters in the
method for Image decomposition. The efficacy of the fusion spatial domain. To retain tumour illumination, increased
is evaluated using Mutual Information (MI) and the contrast is applied to deconstructed CT and MRI Images.
correlation between the fused Image and the reference Second, the three-layer representation uses three fusion
Image. rules.
.
 S. Praveen kumar and S. Sridevi, " Image Fusion  R. Zhu, X. Li, X. Zhang and M. Ma, "MRI and CT
Algorithm for Medical Images using DWT and SR," Medical Image Fusion Based on Synchronized-
2021 International Conference on Artificial Intelligence Anisotropic Diffusion Model," in IEEE Access, 2020:
and Smart Systems (ICAIS), 2021: They proposed a newly developed multi-modal medical
The merged Image will incorporate precise and picture fusion technique centered on the Synchronized-
valuable data from numerous imaging modalities. Discrete Anisotropic Diffusion Equation. (S-ADE). The modified S-
wavelet transform (DWT) is applied to brain region MRI, ADE model, which proves more suitable for MRI and CT
CT, and PET Images. The segments from the low pass and Images, is utilized for decomposing two source Images into
high pass fusion are combined using inverse DWT a single Image. The fusion decision map is computed on
reconstruction. The entropy of the suggested technique is texture layers using the New Sum of Modified Anisotropic
2.9299, which is significantly higher than the state of the art Laplacian (NSMAL) algorithm.
at this time. More accurate and significant information will
be present in the combined Image than in the separate  J. Huang, Z. Le, Y. Ma, F. Fan, H. Zhang and L. Yang,
Images. "MGMDcGAN: Medical Image Fusion Using Multi-
Generator Multi-Discriminator Conditional Generative
 S. A. Begum, K. S. Reddy and M. N. G. Prasad, "CT and Adversarial Network," in IEEE Access, 2020:
MR Image Fusion based on Guided filtering and Phase MGMDcGAN is an adversarial conditional generative
congruency in Non-Subsampled Shearlet Transform network with numerous producers and discriminators. In the
domain," 2021 5th Conference on Information and first cGAN, the generator seeks to create a fused Image that
Communication Technology (CICT), 2021: closely mimics reality by using a specially designed content
Multi-modal Image fusion is essential for the early loss that will fool two discriminators. The discriminators'
diagnosis and management of diseases. To evaluate issues objective is to distinguish structural variations between the
with the human body, medical professionals use imaging fused Image and the source Images. Using this as our
methods like MRI, PET, CT, and SPECT. This is foundation, we employ the second cGAN with a mask to
accomplished using a guided Image filter (GIF) and a Non- enhance the dense structure information in the final fused
subsampled Shearlet transform (NSST) with combined picture without sacrificing functional information. The three
phase congruency-based fusion rules. The GIF gives the fine kinds of medical Image fusion that can be applied using
details needed for the later fusion using the aforementioned MGMDcorgan as a unified strategy are MRI-PET, MRI-
transform technique. This method is tested using MRI and SPECT, and CT-SpECT.
CT scans.
III. EXISTING SYSTEM
 V. A. Rani and S. Lalithakumari, "A Hybrid Fusion
Model for Brain Tumor Images of MRI and CT," 2020 Discrete Wavelet Transform has been used to combine
International Conference on Communication and Signal multimodal medical pictures. This includes Magnetic
Processing (ICCSP), 2020: Resonance Imaging and Computerized Tomography. To
Multimodal medical picture fusion improves functional make up for the information gaps in a single imaging, MRI
and structural information. Fusion may also solve storage and CT are combined to combine the advantages of MRI
issues. An optimal method for a higher-quality hybrid pictures with clear soft tissue information and CT Images'
picture is being researched. This study suggests an empirical clear bone information. An assisted filtering-based MRI and
mode decomposition and discrete wavelet transform-based CT fusion method was proposed by Na et al. (GF). The
multimodal picture fusion paradigm. The recommended fused Image resolves the edge degree and clarity issue by
method creates a combined Image with all of the functional removing the feature data though leaving the edge data of
and spatial details of the original. The combined picture the initial Image untouched. Visual study of the fusion

IJISRT23MAR1668 www.ijisrt.com 2494

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
products' contrary and structural similarity obviously got to combine these two multimodal pictures (as shown in
better. Image 3). Following the wavelet coefficient extraction, a
second fusion algorithm is used to fuse the coefficients.
Discrete wavelet transform has optimal positioning in
the time and frequency domains, which aids in maintaining
the Image's particular information, and it can generate
various input frequency signals while preserving stable
output. The principal component analysis's drawbacks are
overcome by the discrete wavelet transform, which also has
an effect that is both visual and quantitative. The source
Image is enhanced and preprocessed, and the IHS transform,
which retains more anatomical details and lessens color
distortion, is used to extract the intensity component from
the CT Image. The DWT transform is used to create high-
and low-frequency subbands made up of the amplitude
components of CT and MRI. The inverse DWT transform is
used to create the combined Image after the high- and low-
frequency sub bands are fused using separate fusion rules.

Fig 1 CT Scan for Sarcoma

Fig 2 MRI Scan for Sarcoma

The decomposed The decomposed low-frequency

coefficients are merged using the weighted average method,
the high-frequency coefficients are fused using the absolute
high-value method, and the weights are approximated and
optimised using the predator-optimizer, and then the fused
Images are by applying the inverse transform, acquired. Due
to the factors of low and high frequencies have different
meanings, separate fusing rules were used. Therefore, the
idea of fusing is accomplished by using two fusion rules.
First, the most common method of combining features is to
select a larger wavelet coefficient because higher values
indicate stronger edges, which are preferred as an important
component of information content. Two of these important
components of pictures are the corners and borders. Fig 3 Fused Image
Averaging is used to uncover more information about both
of the initial Images, which are approximate based on their The fused Image's mean average accuracy is both the
low wavelet coefficient values. After approximating the lowest and highest when using the discrete wavelet
extracted wavelet coefficients, the merged Image is transform. This is as a result of a geographic resolution
produced using the Inverse Discrete Wavelet Transform. issue. With DWT, an Image is divided into rough and fine
sections, which correlate to lower frequency and higher
Image 1 below displays the CT scan of a patient with frequency sub bands.
sarcoma-related brain illness, and Image 2 displays the MRI
of the same patient. Discrete Wavelet Transform was used

IJISRT23MAR1668 www.ijisrt.com 2495

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Drawbacks of Existing System:  Architecture Diagram:
 Only the spatial connection of the pixels within a single
2-D block is taken into account; the neighbouring blocks'
pixels are not taken into consideration.
 The peak signal to noise ratio is high. Blurring is
produced when bigger DWT basis functions or wavelet
filters are used. This is time consuming process.
 More data set is required for the training purpose, and
also gives less accuracy because it was not done in real
time with 2 input Images
 Therefore has low accuracy and low efficiency in terms
of loading time and implementation time.

 Proposed System:
While ignoring the unwanted and unexpected features,
the complementary information from the source Images
should be incorporated into the final fused Image in an ideal
Image fusing process. The basic idea behind wavelet-based
Image fusion is to combine the two source pictures' wavelet
decompositions using fusion methods on the details and
approximations coefficients.

Without providing functional information, magnetic

resonance imaging (MRI) reveals the soft tissue anatomy of
the brain. A significant role in clinical diagnosis is played by
its great spatial resolution, lack of radiation benefits to the
human body and extensive knowledge. Due to bone's
incredibly low proton density, an accurate MRI bone Image
cannot be created. The bone tissue in the computed  Modules Used:
tomography imaging (CT) Image is especially clear because
bone tissue absorbs density at a greater rate than soft tissue.  Image Decomposition:
Less cartilage information is visible on CT Images, which is In this module, we developed a website, which will
consistent with the anatomical information. Single-photon prompt the user to upload the patient's CT and MRI Images.
emission computed tomography, or SPECT, is a functional The CT and MRI Image will be in the size of 512*512. and
imaging technique that shows the flow of blood in vessels also, we need to give the input as the maximum coordinate,
and veins as well as the metabolism of human tissues and to fix the coordinate in the uploaded source Image for the
organs. It is commonly used in the diagnosis of various purpose of fusion. The uploaded Image is in the form of
tumour diseases and provides both beneficial and RGB. This RGB visual is transformed to a gray scale
detrimental knowledge about tumours. rendition, which has the intensity information. After that
these source gray scale CT and MRI Image is decomposed
Although each will offer particular details, none of using DTCWT.
these methods could deliver all the necessary data in a single
Image. Fusion of medical Images is the answer to this issue.  Extracting Features using DTCWT:
All the necessary details from different sources could be After decomposition we will extract the high intensity
condensed into one Image. and low intensity features as wavelet coefficients. It
produces the various levels of decomposed Images. Each
In order to get around the limitations of the current level the Images are segmented into sub-bands at six
system, the suggested method employs Dual Tree Complex directions like -15˚, -45˚, -75˚, 75˚, 45˚, 15˚ to extract the
Wavelet Transform (DTCWT) to combine various medical features from the Image.
Images. The Discrete Wavelet Transform's shortcomings
will be fixed by the dual tree complex wavelet transform. Principal component analysis is used if the data is
Better directionality and shift variance makes it simple to bigger to reduce the dimension of the data, also extract the
handle the edges and contours of the source picture. The silent and complementary features. As the medical Images
DTCWT's enhanced shift variance and enhanced are larger, to reduce the data here, PCA is used therefore
directionality characteristics make it a successful Image removes the unwanted and repeated information present in
fusion tool. DTCWT. The PCA analysis has several steps. The first is
standardization, the objective of this step is the continuous
initial variables' range needs to be standardised. Therefore,
each will have an equal impact on the research.
Mathematically, standardization is done by finding the
difference between each value and mean and finally

IJISRT23MAR1668 www.ijisrt.com 2496

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
dividing by standard deviation. The next step is to compute in order to invert the transform, and the final step involves
covariance matrix, some variables are highly correlated with averaging the two results.
other variables such a way that they contain redundant
information. Hence, compute the covariance matrix to find This process does not seem to yield a complex
the correlation. To identify the principal components, you transform at all. The transform becomes complex if the real
should also compute the eigenvectors and eigenvalues of the and imaginary components of a complex wavelet are
covariance matrix. Next is to find the featured vector from specified as the outputs of two trees. Odd-length filters have
all the computed principal vectors. Finally, reconstruct the even symmetry around their midpoints while even-length
data with the principal axes. filters have odd symmetry for filters of linear phase PR
biorthogonal sets. These filters' impulse responses resemble
 Image Fusion: the actual and hypothetical components of complicated
The extracted coefficients are approximated. Finally, wavelets. By separable filtration along columns and then
we get the approximated wavelet coefficients. Now the rows, 2-D extension is made possible. The 2-D signal
Image fusion, that is approximated coefficients are fused to spectrum is only maintained in the first quadrant if column
get the fused coefficients. To restore the fused picture, we and row filters reject negative frequencies. Filtering is done
use the Inverse dual tree complex wavelet transform. using complex conjugates of the row filters as well because
two adjacent spectral quadrants should reflect a true 2-D
 Image Segmentation: signal. As a result, the converted 2-D signal has 4:1
The segmentation is used to differentiate the details of redundancy.
lines, curves and boundaries. So the segmentation is
performed on the final fused Image.  Comparitive Analysis:
For regular medical Images, such as CT and MRI
IV. METHODOLOGIES Images, fusion measures are shown in Table 1; for blurred
Image data sets, fusion measures are shown in Table 2. This
The suggested framework enables the use of Dual Tree comparison demonstrates that the proposed approach
Complex Wavelet Transform in the fusion of Images. The outperforms DWT-based Image fusion for both the normal
dual tree complex wavelet transform will be the solution for and blurry medical Image data sets.
the shortcomings of Discrete Wavelet Transform (DWT).
The DTCWT will offer improved directionality and shift Table 1 Performance Measure for Normal Image Data Set
variance, making it simple to process the source Image's
edges and contours. Increased shift variance and increased
directionality features of the DTCWT make fulfilled Image
fusion tool. Medical imaging systems are helpful to detect
various defects and issues in our body. X-rays, Positron
Emission Tomography (PET), Single Photon Emission
Computed Tomography, Computed Tomography, Magnetic
Resonance Imaging (MRI), and Computed Tomography
(SPECT) are some types of medical imaging system. Each
will provide unique details, but none of those applications
could give all the information required in a single Image.
Fusion of medical Images is the answer to this issue. One
picture could contain all the pertinent data from various
source Images. Table 2 Performance Measure for Blurred Image Data Set
Dual Tree Complex Wavelets could provide reasonable
shift invariance and good directionality, but using a single
tree it was impossible to achieve perfect reconstruction and
acceptable frequency characteristics. However, approximate
shift invariance could be obtained by tripling the sampling
rate at each tree level. Samples must be equally spaced for
it. This could be acquired by eliminating down sampling of
2 after level 1 filters, H0a and H1a. Another way to
accomplish this, two parallel completely decimated trees, a
and b, will be used. Under level 1, filters from one tree
should have a half-sample delay distinct from filters from
another tree in order to achieve uniform intervals between
samples of the two trees. This requires odd-length filters in
one tree and even-length filters in another tree in order to
achieve linear phase. Greater symmetry exists between trees
if even and odd filers are used in levels of tree, alternately.
PR filters are used as usual to invert each tree independently

IJISRT23MAR1668 www.ijisrt.com 2497

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. RESULTS [2]. Padmavathi K, Maya V KarKi, Mahima Bhat,
“Medical Image Fusion of Different Modalities
In addition to research showing that the Dual Tree Using Dual Tree Complex Wavelet Transform With
Complex Wavelet Transform (DTCWT) gives the fused PCA”, 2016 International Conference on Circuit,
Image a more accurate representation of the spectral, spatial, Communication and Computry (I4C), October 2022.
and soft tissue details of the tumor, it is the basis for our [3]. Rajiv Singh, Richa Srivastava, Om Prakash and
proposed approach of medical Image fusion. Our suggested Ashish Khare, “Mixed scheme based multimodal
approach has high entropy, fusion factor, and peak signal to medical Image fusion using Daubechies Complex
noise ratio values with regard to performance level. Because Wavelet Transform”, IEEE/OSA/IAPR International
of this, our suggested approach outperforms other fusion Conference on Informatics, Electronics & Vision,
techniques. (ICIEV), pp. 304-309, 2022.
[4]. Negar Chabi, Mehran Yazdi, Mohammad
VI. CONCLUSION Entezarmahdi, “An efficient Image fusion method
based on dual tree complex wavelet transform”, 8th
Disease diagnosis and therapy can both benefit from IEEE Iranian Conference on Machine Vision and
medical Images. The combined medical Image will be used Image Processing (MVIP), pp. 403-407, 2021.
to make more precise diagnoses and collect detailed data for [5]. Ivan W. Selesnick, Richard G. Baraniuk, Nick G.
improved treatment. However, we are not getting more Kingsbury, “The Dual-Tree Complex Wavelet
precise pictures from the fused. to produce a fused picture Transform, A coherent framework for multiscale
that is visually intensified. In this endeavor, we signal and Image processing”, IEEE, signal
demonstrated dual-tree complex wavelet transform-based processing Magazine, pp.123-151 Nov-2021.
medical fusion. [6]. Himanshi, Vikrant Bhaterja Abhinav Krishn, and
Akanksha Sahu, “An improved Medical Image
The fused multi-model picture will be calculated using Fusion Approach Using PCA and Complex
a dual-tree complex wavelet transform and will exhibit Wavelets”, IEEE International conference on Medical
improved shift variance and directionality selectivity. The Imaging, m- Health and Emerging Communication
source Images used in this project, such as CT and MRI systems (MedCom) , pp. 442-447, 2021.
scans, are accepted, and our suggested technique [7]. Sruthy S, Dr. Latha Parameswaran , Ajeesh P Sasi, “
decomposes them using DTCWT. Additionally, coefficient Image Fusion Technique using DT-CWT”, IEEE, pp.
extraction and estimate of extracted coefficients are carried 160-164, 2021.
out. The merged Image is then obtained by applying the [8]. Sonali Mane, S.D. Sawant, “ Image Fusion of
inverse dual tree complex wavelet transform, and the CT/MRI using DWT, PCA Methods and Analog DSP
segmented fused Image is then obtained by segmentation. Processor”, International Journal of Engineering
As a result, the fused picture of our suggested scheme shows Research and Applications ISSN:2248-9622, Vol. 4,
the tumour's more precise edges, tissues, spectral, contour, Issue 2(Version 1), pp.557-563, February 2021.
and spatial features. [9]. Rajiv Singh, Ashish Khare, “Multimodal medical
Image fusion using Daubechies complex wavelet
 Scope for Future Development: transform”, Proceedings of IEEE Conference on
The technique we suggest of medical Image fusion is information and communication technologies (ICT)
based on Dual Tree Complex Wavelet Transform because ,pp. 869-873, 2020.
study indicates that it provides the fused Image a more [10]. P.R.Hill, D.R.Bull & C.N Canagarajah, “ Image
accurate representation of the spectral, spatial, and soft tissue fusion using a new framework for complex wavelet
details of the tumour. (DTCWT). Our proposed method transform”, IEEE, September, 2020.
performs well in terms of peak signal to noise ratio, fusing [11]. J. Ruf, E. Lopez Hänninen, M. Böhmig, I. Koch, T.
factor, and entropy. As a result, when compared to other Denecke, M. Plotkin, et al. “Impact of FDG-
fusion techniques, our suggested algorithm works well. PET/MRI Image fusion on the Detection of
Block level fusion can be used in future work, and the Pancreatic Cancer” . pancheratology , vol.6, pp.512-
outcomes, including performance assessment and the end 519, 2020.
fused Image regarding enhanced visual representation, can [12]. Jambor, R. Borra, J.Kemppainen, V.Lepomäki,
be further enhanced. R.Parkkola, K. Dean, et al.” Improved detection of
localized prostate cancer using co-registered MRI and
REFERENCES 11C-acetate PET/CT” , Europian Journal of
Radiology, vol.81 , pp. 2966-2972, 2020.
[1]. Tannaz akbarpour, Mousa shamsi, Sabalan [13]. S.Kor,U.S.Tiwary,”feature level fusion of multimodal
Daneshvar, “Structural medical Image fusion by medical Images in lifting wavelet transform domain”,
means of dual tree complex wavelet”, The 22nd Engineering in Medicine and biology Society,IEEE
Iranian Conference on Electrical engineering (ICEE 2020
2014), pp.1970- 1975, May 20-22, 2022. [14]. I.W.Selesnick, R.G.Baraniuk, N.C.Kingsbury. “ The
dual tree complex wavelet transform “ . Signal
processing magazine,vol.22,pp. 123-151,2020.

IJISRT23MAR1668 www.ijisrt.com 2498

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[15]. J.Magarey, N.C.Kingsbury. “ Motion estimation using
a complex- valued wavelet transform”. Signal
Processing , IEEE transactions on, vol.46,pp.1069-
1084,2020.

IJISRT23MAR1668 www.ijisrt.com 2499