Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract The purpose of an image fusion for medical images is to associate a num-
ber of images gained from many bases to a solitary image appropriate for better
analysis. The vast majority of the best in class image fusing systems are based on
non-fuzzy sets, and the fused image so obtained lags with complementary infor-
mation. Fuzzy sets are strong-minded to be more appropriate for medical image
processing as more hesitations are considered compared with non-fuzzy sets. In this
paper, a procedure for efficiently fusing multimodal medical images is presented. In
the proposed method, images are initially converted into intuitionistic fuzzy images
(IFIs), and another target work called intuitionistic fuzzy entropy (IFE) is utilized
for membership and non-membership capacities to accomplish the finest estima-
tion of the bound. Next, the IFIs are compared using the fitness function, entropy.
Then, orthogonal teaching–learning-based optimization (OTLBO) is introduced to
optimize combination factors that change under teaching phase, and learner phase of
OTLBO. Finally, the fused image is achieved using optimal coefficients. Simulations
on several pairs of multimodal medical images are performed and matched with the
current fusion approaches. The dominance of the proposed technique is presented
and justified. Fused image quality is also verified with various quality metrics, such
as peak signal-to-noise ratio (PSNR), universal quality index (UQI), structural simi-
larity (SSIM), correlation coefficient (CC), entropy (E), spatial frequency (SF), edge
information preservation (QAB/F ), and standard deviation (SD).
1 Introduction
Multimodal medical image fusion (MMIF) [1] is the progression of bonding two mul-
timodal medical images to increase the quality of output image. In merging medical
image pairs, such as computed tomography–magnetic resonance imaging (CT–MRI),
magnetic resonance imaging–magnetic resonance angiography (MRI–MRA), X-
ray–vibro-acoustography (X-ray–VA) [2], MRI–PET (magnetic resonance imag-
ing—positron emission tomography) [3], and MRI–SPECT (magnetic resonance
imaging–single photon emission computed tomography), image fusion is gradually
achieving consequence in healthcare and modern medicine. These sets of images
are used for mining clinical facts that are complimentary in nature. For example,
CT image ambiances with a lesser amount of distortion and conveys details regard-
ing condensed structures like bones. MRI offers neurotic soft fleshy tissue evidence
and MRA intellects easily brain defects. X-ray identifies fractures and irregularities
in bone position while VA provides depth and width of a disease object. PET and
SPECT provide functional and metabolic information of the human brain. Hence, all
significant data cannot be obtained in a single image and henceforth MMIF is essen-
tially necessary to achieve all probable comprehensive data in a single combined
image called fused image.
Fuzzy sets play a substantial role in image handling to reduce the obscurity and
ambiguity existing in images. Ambiguity in fuzzy sets is taken as a membership
function which deceits the interim [0, 1] where 0 characterizes no membership and 1
characterizes full membership. Fuzzy set theory anticipated by Zadeh [4] in 1965 is
not able to directly model uncertainties. Atanassov [5] in 1986 presented an intuition-
istic fuzzy set (IFS), a general form of fuzzy set hypothesis. IFS takes two ambiguity
parameters—the membership degree and non-membership degree (due to the hesi-
tation degree). Real-time complications with indefinite understanding are resolved
effectively by the non-membership degree. In each period of image handling, numer-
ous vulnerabilities happen. These vulnerabilities are reduced by means of IFSs, and
the image is boosted in terms of contrast of the image.
Rao et al. [6] formerly advanced a universal optimization process called
teaching–learning-based optimization (TLBO). The TLBO shows nearly common
features of former evolutionary computation (EC) procedures which are a population-
based iterative learning process. In any case, TLBO analyzes for an optimum over
every learner attempting to accomplish the understanding of the teacher. TLBO
attains optimum outcomes and is preserved as the best-learned individual in the soci-
ety than through learners experiencing genomic processes like mutation, crossover,
and selection. TLBO is successfully applied to many real-world problems [7, 8] due
to its modest idea and high proficiency, and it emerged as a very pretty optimization
technique. In general, TLBO yields better results to the other EC methods.
In this paper, an orthogonal design process is included in the elementary TLBO
optimization. In this procedure, every student (learner) over the span of students in
the orthogonal design is isolated into a few partial vectors where each of the stu-
dents goes about as a factor. Orthogonal design, among all the different groupings, is
Multimodal Medical Image Fusion Based on Fuzzy Sets … 489
included to look through the finest scales. An orthogonal design process [9] demon-
strates with orthogonal array (OA) and factor, however, an illustrative arrangement
of courses of action for research to achieve respectable measures. The array is called
orthogonal in light of the fact that all sections are assessed freely of each other, and
the primary result of one factor does not inconvenience the evaluation of the funda-
mental impact of another factor. Experimental results proved that this technique is
superior to the performance of the traditional EAs. To make orthogonal design faster
and robust, TLBO is applied to an orthogonal design which is called OTLBO [10]
and is implemented in this paper.
2 Preliminaries
Medical images are poorly illuminated, contain many uncertainties in the form of
noise, having vague boundaries, gray levels are overlapped, invisible blood ves-
sels, and difficult to extract objects from the image. Fuzzy sets in image processing
increase contrast, flat the areas of interest, and refine the boundaries and fine erections
of the image.
Atanassov also introduced a novel constraint πI (y) called the hesitation degree or
intuitionistic fuzzy index, rising due to the absence of information or individual fault
in conveying the membership degree. An IFS, I in a finite set Y with πI (y) + μI (y) +
vI (y) 1, and the introduction of the hesitation degree may be represented as
490 T. Tirupal et al.
n
IFE(Z; α) πZ (yi ) exp(1 − πZ (yi )) (4)
i1
where πZ (yi ) 1 − (μZ (yi ) + vZ (yi )) denotes the hesitation degree, μZ (yi ) denotes
the membership degree, and vZ (yi ) denotes the non-membership degree.
IFE utilizes Eq. (4) for designing and takes α standards going [0.1–1]. The extreme
entropy standards are calculated by α optimization.
The membership degrees of the IFIs are found by utilizing the known α values,
and IFI is obtained as
2.2 TLBO
Step 4: Compare the solutions, if the new solution is not superior to present, then
retains the former solution.
Step 5: Select two solutions randomly, on the off chance that the new arrangement
is superior to existing. At that point, another arrangement is discovered
utilizing.
⎧
⎨ X l + rand ∗ (X l − X l ) if (f (X l ) < f (X l ))
m m n (i) (r)
l
Xnew(i) (8)
⎩ X l + rand ∗ (X l − X l ) otherwise
m n m
Step 6: The obtained new solution is superior to the previous solution and is taken
as the best solution otherwise the previous solution is the best solution.
Step 7: After satisfying termination criteria, the final optimal coefficients are
obtained. After completing the maximum number of iterations the algo-
rithm is terminated.
2.3 Entropy
L
H − hIf (i) log2 hIf (i) (9)
i0
where hIf (i) speaks to the standardized histogram of the combined image and L means
the measure of recurrence receptacles in the histogram.
Fuzzification Fuzzification
Decomposition Decomposition
OTLBO
where Iij1 characterizes the dim level of the primary image and assortments from
0 to L − 1 (L is the outrageous dark-level esteem). lmin and lmax speak to the
base and most extreme dark-level estimations of the primary image.
(3) Calculate the optimum value of α using entropy Eqs. (4) and (5) for the first
image and this α varies for different input images.
(4) With the optimum value of α, find fuzzified IFI for the input image I1 using
beneath equations and indicate as IY 1 .
where vIFI 1 (Iij1 ; α) is the optimized non-membership degree of the first image.
where πIFI 1 (Iij1 ; α) is the optimized hesitation degree of the first image.
IY 1 (Iij1 , μIFI 1 (Iij1 ; α), vIFI 1 (Iij1 ; α), πIFI 1 (Iij1 ; α)) (14)
(5) Repeat the above procedure from step 2 to step 4 for the second input image, I2
to find IY 2 .
IY 2 (Iij2 , μIFI 2 (Iij2 ; α), vIFI 2 (Iij2 ; α), πIFI 2 (Iij2 ; α)) (15)
where μIFI 2 (Iij2 ; α), vIFI 2 (Iij2 ; α), πIFI 2 (Iij2 ; α)) are optimized membership,
non-membership and hesitation degrees of the second image, respectively.
(6) Images IY 1 and IY 2 are decayed into m × n windows, and in this algorithm, 5 × 5
is the window size taken and then the entropy of each block of two images is
computed separately.
(7) Then, perform OTLBO algorithm to find optimal coefficients.
(8) Lastly, utilizing these optimal coefficients, the final fused image is attained, and
the fused image is filtered for noise removal.
Multimodal Medical Image Fusion Based on Fuzzy Sets … 495
Table 1 Performance comparison of fused images of TLBO algorithm [8] and the proposed OTLBO algorithm
Modality pair Fusion PSNR UQI SSIM CC E SF QAB/F SD
technique
CT–MRI TLBO [8] 61.23 0.762 0.912 0.913 6.294 21.442 0.711 46.85
OTLBO 65.18 0.798 0.998 0.948 6.703 27.29 0.859 90.76
MR–MRA TLBO [8] 64.33 0.667 0.783 0.814 5.926 25.73 0.631 49.45
OTLBO 69.73 0.823 0.846 0.892 6.968 41.36 0.751 79.44
MRI–PET TLBO [8] 62.16 0.776 0.801 0.835 6.273 43.18 0.625 50.37
OTLBO 69.93 0.784 0.824 0.863 6.635 44.24 0.767 70.64
MRI–SPECT TLBO [8] 65.78 0.48 0.841 0.863 5.595 35.63 0.408 21.57
OTLBO 67.49 0.771 0.935 0.875 6.618 43.56 0.673 52.64
X-ray–VA TLBO [8] 61.28 0.714 0.813 0.837 6.796 39.51 0.775 39.02
OTLBO 67.69 0.882 0.915 0.872 7.185 53.64 0.872 48.82
PET–CT TLBO [8] 65.75 0.781 0.889 0.874 6.268 35.49 0.734 43.08
OTLBO 67.63 0.855 0.945 0.887 6.765 44.65 0.758 83.59
T. Tirupal et al.
Multimodal Medical Image Fusion Based on Fuzzy Sets … 497
1.
2.
3.
4.
5.
6.
Fig. 3 Fusion results of multimodal medical images using TLBO algorithm [8] and the proposed
OTLBO algorithm
than that of existing method. It has a high average number of bits per pixel of 7.185
for the proposed method which gives more data content than the current technique.
The last case speaks to multimodality images, for example, the PET and CT
image appeared in Fig. 3-6a, b. The PET picture shows the cerebrum work, has a low
spatial assurance. A CT picture gives thick structures like bones and embeds with less
bending; it cannot recognize practical changes. Intertwining these two pictures, we
get both practical information and extra spatial highlights without spatial mutilation
and this is found in Fig. 3-6d for the developed algorithm.
5 Conclusions
This paper presents the optimal block size for the fusion of multimodal medical
images by a global optimization procedure called OTLBO. The optimum solution
is achieved by an efficient search of a large sample space using orthogonal design.
Here, for a fixed number of iterations, algorithm is run and optimal coefficients
are obtained. So, finally the final fused image is acquired. The obtained results are
compared with existing TLBO with respect to different quantitative measures. From
the result analysis, it is obvious that OTLBO outstrips the existing method in terms
of accuracy. We can achieve that OTLBO is an intense technique in advancing the
blocks that are discrete and multimodal in giving quality ideal outcomes.
References
1. Yang, Y., Park, D.S., Huang, S., Rao, N.: Medical image fusion via an effective wavelet based
approach. EURASIP J. Adv. Signal Proc. (2010)
2. Gholam, H.H., Alizad, A., Fatemi, M.: Integration of Vibro-acoustography imaging modality
with the traditional mammography. Int. J. Biomed. Imaging. Hindawi Publishing Corporation
(2007)
3. Baum, K.G., Raerty, K., Helguera, M., Schmidt, E.: Investigation of PET/MRI image fusion
schemes for enhanced breast cancer diagnosis. In: Proceedings of IEEE Seventh Symposium
Conference on Nuclear Science (NSS) (2007)
4. Zadeh, L.A.: Fuzzy sets. Information. Control 8, 338–353 (1965)
5. Atanassov, K.T.: Intuitionistic fuzzy sets. Fuzzy Sets Syst. 20, 87–96 (1986)
6. Rao, R.V., Savsani, V.J., Vakharia, D.P.: Teaching-learning-based optimization: a novel method
for constrained mechanical design optimization problems. Comput. Aided Des. 43(3), 303–315
(2011)
7. Vedat, T.: Design of planar steel frames using teaching-learning based optimization. Eng. Struct.
34, 225–232 (2012)
8. Jin, H., Wang, Y.: A fusion method for visible and infrared images based on contrast pyramid
with teaching learning based optimization. Infrared Phys. Technol. 64, 134–142 (2014)
9. Fang, K.T., Ma, C.X.: Orthogonal and Uniform Design. Science Press, Beijing (2001)
10. Suresh, C.S.: A teaching learning based optimization based on orthogonal design for solving
global optimization problems. Springer Plus 2, 130 (2013)
11. Chaira, T.: A novel intuitionistic fuzzy c means clustering algorithm and its application to
medical images. Appl. Soft Comput. 11(2), 1711–1717 (2011)
Multimodal Medical Image Fusion Based on Fuzzy Sets … 499
12. Jagalingam, P., Hegde, A.V.: A review of quality metrics for fused image. In: Aquatic Procedia
of International Conference on Water Resources, Coastal and Ocean Engineering (ICWRCOE)
(2015)
13. Tirupal, T., Mohan, B.C., Kumar, S.S.: Multimodal medical image fusion based on sugeno’s
intuitionistic fuzzy sets. ETRI J. 39(2), 173–180 (2017)
14. www.med.harvard.edu/aanlib
15. www.metapix.de
16. www.mathworks.in