Hindawi

Applied Bionics and Biomechanics

Volume 2022, Article ID 2047537, 11 pages
https://doi.org/10.1155/2022/2047537

Research Article
Implementation of CT Image Segmentation Based on an Image
Segmentation Algorithm

Lingli Shen
Shanghai East Hospital, Shanghai, China

Correspondence should be addressed to Lingli Shen; [email protected]

Received 2 August 2022; Revised 8 September 2022; Accepted 20 September 2022; Published 12 October 2022

Academic Editor: Ye Liu

Copyright © 2022 Lingli Shen. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

With the increasingly important role of image segmentation in the field of computed tomography (CT) image segmentation, the
requirements for image segmentation technology in related industries are constantly improving. When the hardware resources
can fully meet the needs of the fast and high-precision image segmentation program system, the main means of how to
improve the image segmentation effect is to improve the related algorithms. Therefore, this study has proposed a combination
of genetic algorithm (GA) and Great Law (OTSU) algorithm to form an image segmentation algorithm-immune genetic
algorithm (IGA) algorithm. The algorithm has improved the segmentation accuracy and efficiency of the original algorithm,
which is beneficial to the more accurate results of CT image segmentation. The experimental results in this study have shown
that the operating efficiency of the OTSU segmentation algorithm is up to 75%. The operating efficiency of the GA algorithm
is up to 78%. The operating efficiency of the IGA algorithm is up to 92%. In terms of operating efficiency, the OTSU
segmentation algorithm has more advantages. In terms of segmentation accuracy, the highest accuracy rate of OTSU
segmentation algorithm is 45%. The accuracy of the GA algorithm is 80%. The highest accuracy of the IGA algorithm is 97%.
The IGA algorithm is more powerful in terms of operating efficiency and accuracy. Therefore, the application of the IGA
algorithm to CT image segmentation is beneficial to doctors to better judge the lesions and improve the diagnosis rate.

1. Introduction the images, so as to evaluate the health status of the human

body. In addition, radiation CT, digital stereography, ultra-
With the improvement of people’s quality of life, the sound imaging, magnetic resonance imaging, and modern
society’s demand for the medical industry is increasing. In medical diagnosis have undergone major changes due to
order to simplify the work of doctors’ diagnosis, computer- computer and medical image processing techniques that
aided diagnosis technology has been widely researched and have laid the foundation for the development of these imag-
developed. The abdomen is one of the parts of the human ing techniques.
body with many tissues and organs, the most complex struc- Technology for medical diagnosis and treatment has
ture and high incidence. Therefore, the diagnostic technol- advanced significantly, and several novel CT image segmen-
ogy of abdominal medical computed tomography (CT) tations are suitable for clinical diagnosis and therapy. Addi-
images has important research value. How to accurately tionally, the data generated by combining several imaging
locate and segment organs and tissues from medical images modalities can reach complementarity and offer a solid
is an important prerequisite for the realization of abdominal scientific foundation for biomedical research and clinical
medical diagnosis technology. The area of medical imaging diagnostics. As a result, CT image segmentation technology
underwent a revolution when CT technology was developed has received a lot of attention and has advanced quickly.
in the early 1970s. Medical imaging shows the structure and Medical image processing has gotten more and more atten-
density of the internal tissues and organs of the human body tion from academics as CT image segmentation technology
in the form of images, which is used by the diagnosing phy- has continued to advance. The most important technique
sicians to judge according to the information provided by for processing and analyzing medical images is CT image
2 Applied Bionics and Biomechanics

segmentation. In image medicine, picture segmentation is tation, and they have highlighted the great promise of con-
becoming more and more crucial. Image segmentation is a volutions using upsampling filters. As a powerful tool in
crucial tool for removing quantitative data from specific tis- dense prediction tasks, convolution can control the resolu-
sues in images. It is also a necessary step before visualization tion in deep convolutional neural networks [6]. Mariano
and is applied in many different fields. The innovation of et al. has proposed an image segmentation framework based
this study is that the basic algorithm based on the image seg- on maximum likelihood estimation, which has been gener-
mentation algorithm is proposed, and the image segmenta- ated by sampling with different likelihood functions. They
tion algorithm is improved based on the genetic algorithm have also proposed efficient segmentation methods capable
(GA). Experiments have proved that the proposed algorithm of dealing with textured images [7]. A fresh heterogeneous
is meaningful. neural network model for image segmentation has been
put out by Yang et al. The image is divided into many sec-
2. Related Work tions based on the gray level, and different neurons correlate
to various parameters. The outcomes have proven that the
One of the medical technology development fields with the suggested segmentation strategy is successful and efficient
quickest growth rate over the past 20 years is medical CT [8]. Mean shift, a widely used algorithm for multi-feature
imaging. This technology enables clinicians to observe the segmentation, has been discovered by Cuevas et al. to be
internal lesions of the human body more directly and clearly, more effective than one-dimensional feature methods. They
and the diagnosis rate is also improved. Suzuki et al. has have developed a novel, competitive segmentation technique
found that one needs to evaluate not only the value of for grayscale images that has already taken into account the
mechanical properties but also the degree of damage in local variance and grayscale value of each pixel [9]. Scholars
order to maintain concrete structures effectively. They have have proposed that neural network can be used in image seg-
suggested the use of X-ray CT for damage estimation of con- mentation to improve the accuracy and efficiency of image
crete [1]. Ha et al. has found that the attribution of reduced segmentation, and it has been considered that image seg-
intracranial cerebrospinal fluid (CSF) volume in patients mentation can achieve important development in various
with mass-occupying masses is associated with reduced cere- fields. However, scholars have not presented specific cases
bral perfusion pressure and subsequent cerebral ischemia. to prove this point.
Their aim was to assess the ratio of CSF volume to brain vol-
ume using CT images [2]. Dijk et al. has found that CT 3. CT Image Segmentation Method Based on
images currently used to predict moderate-to-severe xero- Image Segmentation
stomia in advanced post-radiotherapy patients are highly
accurate. They have investigated 249 patients with head With the emergence of computer-aided diagnosis and med-
and neck cancer who received definitive CT scans and radi- ical images for clinical auxiliary diagnosis, the diagnosis effi-
ation therapy [3]. Zorn et al. has been designed to build the ciency of clinicians has been greatly improved, and this field
learning curve associated with achieving standardized post- has attracted more and more attention of researchers. In the
processing for radiographer students. To discuss their inter- realm of image processing and computer vision, picture seg-
ests, each student has repeated 2 of the 15 CT post- mentation is a crucial component and a challenge [10–12].
processing protocols. They have also assessed each student’s The technology and procedure of dividing an image into sec-
achievement accuracy at each repetition. Results have shown tions with various properties and extracting items of interest
that the learning rates for the two protocols are 63% and is known as image segmentation. This is both a fundamental
56%, respectively [4]. Li et al. has found that norm regulari- computer vision technique as well as a crucial step from
zation has attracted attention in computerized CT scan image processing to image analysis. It makes it possible to
image reconstruction. CT image gradients have provided a grasp feature extraction and parameter measurement targets
measure of image gradient sparsity for reconstructing at a higher level of picture analysis. Research on image seg-
images from finite projection data in CT and have effectively mentation techniques is crucial since the subsequent pro-
addressed optimization problems [5]. Scholars have seen cessing directly depends on how accurate the segmentation
that CT images are not only used in the medical field, but results are.
also used in various fields to a certain extent, which also CT is less effective for soft tissue contrast imaging, but
shows that CT images have made great progress. However, with high density resolution, it can clearly display bone,
applying image segmentation to it would make it more muscle, and organ information [13]. CT scans a certain part
effective. of the human body one by one with a highly sensitive detec-
Medical image segmentation is the process of dividing an tor. It has the characteristics of fast scanning time and clear
image into several regions according to the similarity within image. Each scan needs to inject a contrast agent, which is a
the region and the difference between the regions. How to scan image with radioactive radiation, the cheapest price,
separate relevant structures and regions of interest from and the most widely used medical imaging. Magnetic reso-
images is the first problem to be solved by image analysis nance imaging is an imaging based on radio waves and has
and recognition, and it is also a bottleneck that limits the the highest spatial resolution. It has better resolution for soft
development and application of other related technologies tissue (cranial structure) in the human body and can achieve
in medical image processing. Chen et al. has discovered that functional imaging. However, it has low sensitivity, long
deep learning can be used to solve semantic image segmen- image acquisition time, and moderate price [14]. However,
Applied Bionics and Biomechanics 3

Figure 1: CT imaging.

since only labeled CT images are currently published in the the P-tile method [17]. The P-tile algorithm is an automatic
field of liver segmentation, CT imaging is used under data threshold selection algorithm based on gray histogram sta-
constraints. CT imaging is shown in Figure 1. tistics, which needs to be based on certain prior conditions.
As shown in Figure 1, the cross-sectional images The schematic diagram of image segmentation is shown in
obtained by CT scanning can also be reorganized and recon- Figure 2.
structed by computer software to obtain multi-plane cross- As it can be seen from Figure 2, the existing image seg-
sectional images such as sagittal and coronal planes required mentation methods are mainly divided into the following
for diagnosis. The CT medical device scanning apparatus categories: threshold-based segmentation methods, region-
sends X-rays into the body’s numerous tissues and organs, based segmentation methods, edge-based segmentation
returning varying response values depending on the organs methods, and specific theory-based segmentation methods.
and tissues. A particular interface function is needed to read In the 1980s, there was a theory based on the extraction of
medical data in DICOM format files, which are the data out- full-image pixel information combined with mathematical
put from CT medical equipment, in order to load images methods to obtain the best fitness. The reason why this the-
[15]. Diverse CT scanning imaging equipment is employed ory is not widely accepted is that its noise reduction is not
as a result of the diversity of CT medical scanning equip- ideal and the workload is large [18]. It is expressed as for-
ment, and the gray value range of the CT image produced mula (2):
by the corresponding scanning is also different [16]. Setting
the appropriate window width and window level values for s t

the CT image is required in order to display the pixel range w0 ðs, t Þ = 〠 〠 Pi, j : ð2Þ
i=0 j=0
truncation and display the DICOM data image appropri-
ately. The values within the window width are mapped to
the values between [0, 255] through linear transformation, The OTSU algorithm is a preliminary progress in the
defined as formula (1): processing of pixel extraction gray histogram in image seg-
mentation. After obtaining the gray level information of
  the histogram, the optimal value is obtained by comparing
V − ðC − 0:5Þ
GðV Þ = + 0:5 × 255:0, ð1Þ the algorithm, and the important operator of image segmen-
W −1 tation—threshold value is determined from this [19]. OTSU
is a commonly used algorithm in threshold segmentation,
where 255.0 is the original image data, W is the window which can automatically generate the best segmentation
width value, C is the window level value, and GðVÞ is the threshold based on images. The core idea of OTSU is to
converted pixel value. maximize the variance between classes. The OTSU algo-
rithm does not need any manual evaluation work in the
3.1. Image Segmentation Based on GA. In the 1960s, some early stage, but the process of obtaining the optimal solution
scholars proposed an operation that conformed to the hard- requires a huge number of operations. If this contrast is
ware conditions at that time. The most obvious disadvantage defined as a matrix, the important variable of the operational
of this automatic threshold selection method based on gray formula is as follows formula (3):
histogram is that it needs to measure the ratio of the image
segmentation target to the whole image before segmentation. s t
The artificial measurement method results in a very low μi ðs, t Þ = 〠 〠 iPi : ð3Þ
repetition rate of this segmentation operation, known as i=0 j=0
4 Applied Bionics and Biomechanics

Figure 2: Schematic diagram of image segmentation.

evaluation selection

mutation

crossover

Figure 3: The structure diagram of the genetic algorithm.

When the variance value is formula (4), the minimum better optimization results faster. GAs have been widely used
gray value to the maximum gray value are involved. in combinatorial optimization, machine learning, signal pro-
cessing, adaptive control, and artificial life. The basic idea of
s t GA is to mathematically describe Darwin’s theory of biolog-
μ j ðs, t Þ = 〠 〠 jP j : ð4Þ ical evolution, so as to abstract it for modeling and form a
i=0 j=0 traversal algorithm. GA has strong and comprehensive par-
allelism in calculation, which can reasonably prevent prema-
From the analysis of the OTSU algorithm, it can be ture convergence, and parallelism can be effectively hidden
known that w0 ðs, tÞ, μi ðs, tÞ, and μ j ðs, tÞ should not only or contained in the process of crossover and mutation calcu-
accumulate and sum, but also incorporate all the summation lation. The noise processing method is relatively mild, and
results into all (s, t). After the summation and merging of has the advantages of certain elimination and identification
huge data, this operation is very complex and time- measures and reasonable calculation speed. The structure
consuming [20]. diagram of the GA is shown in Figure 3.
In order to reduce the time complexity, some scholars As shown in Figure 3, GAs generally do not need other
use efficient hardware equipment with specific calculation external information in the process of search evolution,
principles to calculate the OTSU algorithm, and the effect and only use evaluation functions to evaluate the pros and
is also not very ideal [21]. Moreover, once the gray value dis- cons of individuals or solutions, thus serving as the basis
tribution of the segmented source image is chaotic, and the for subsequent genetic operations. Using GA combined with
normal extreme value cannot be obtained, the high- the idea of region segmentation technology in image seg-
precision segmentation result cannot be obtained by using mentation theory, the image to be segmented can be seg-
the OTSU method. mented and modeled. The optimal threshold can be
With the continuous in-depth research of previous obtained by performing modeling operations in the model,
scholars, a theoretical idea that analyzes the solution of and this threshold is the segmentation result that can be
image segmentation operators with evolutionary theory has obtained in the image area that needs to be segmented.
emerged, that is, the early GA is applied to the image seg- Because mathematical modeling is very difficult, how to
mentation threshold [22]. Compared with some conven- use the mathematical model to obtain the best threshold to
tional optimization algorithms, GA can usually obtain obtain the ideal segmentation effect has always been the
Applied Bionics and Biomechanics 5

difficulty of image processing technology, especially image get information during the calculation process, the
segmentation technology. In this study, it tries to combine segmentation effect may be deviated. Therefore, it is neces-
the improved GA optimization method with the maximum sary to use the algorithm to keep the target away from the
inter-class variance method improved by this study, and background, that is, the larger the class spacing between
gives full play to the advantages of each method to achieve the two, the noise would be shielded, so that the segmented
a good segmentation effect. In the GA calculation, in order target would be more accurate.
to make the algorithm retain the individual’s good genes, The binarization of the image is to set the gray value of
the adaptive GA can be used to adjust the traditional GA the pixels on the image to 0 or 255, that is, the entire image
adaptive degree problem. Let the crossover operator be Pc, presents an obvious visual effect of only black and white. At
if they need to be adjusted automatically with the fitness, the same time, after the image is binarized, the distance
there is formula (5): between each piece of pixel information and the image cen-
ter information is opposite. The closer the distance is, the
8  
>
> k1 f max − f ′ better the cohesion of the pixels between each category is.
< , f ≥ f avg This idea uses the average variance of the two types to
Pc = f max − f avg : ð5Þ express as formula (8):
>
>
:
k2 , f < f avg
1
σ 0 2 ðt Þ = 〠 ði = μ0 ðt ÞÞ2 pðiÞ: ð8Þ
Among them, f max is the largest fitness value in a specific w0 ðt Þ 0≤t
group. favg is the average fitness value of each generation of
the group. f ′ is the larger fitness value in the operator that
In this formula, μ0 ðtÞ and w0 ðtÞ represent the target gray
requires crossover operation. f is the general fitness value
value and the background gray value, respectively, and t is
that requires mutation operation. These coefficients are
the gray threshold. Through variance operation, it can be
often excellent seeds selected in the early stage of evolution,
obtained that the more uniform the distribution of pixels
and they remain unchanged when other seeds evolve, thus
within the class, the better the cohesion between individuals,
shielding their own evolution and causing the calculus to fall
so that the target and the back are easier to distinguish.
into a local optimum.
In order to solve the problems of class cohesion and the
Adding two variables f max is the maximum mutation
distance between the target and the background class, this
probability and Pc1 is the maximum crossover probability.
study improves the threshold value based on the basic OTSU
It is expressed as formula (6):
method is defined as formula (9):
8 
< P ⋅ e f −f avg f max −f avg , f ≥ f
Pc = c1 avg
: ð6Þ w0 ðt Þw1 ðt Þðμ0 ðt Þ − μ1 ðt ÞÞ2
:P , f < f G ðt Þ = : ð9Þ
c1 avg σ20 + σ21

The so-called activation function is the function running

When G(t) takes the maximum value, the corresponding
on the neurons of the artificial neural network, which is
gray level is the obtained optimal threshold value Th. It is
responsible for mapping the input of the neuron to the out-
expressed as formula (10):
put. In a neural network, there is a neuron activation func-
tion sigmoid that can balance linear and nonlinear
behavior, and its expression is as shown in formula (7). Th = arg max ½Gðt ފ: ð10Þ

1
ϕðvÞ = : ð7Þ GðtÞ and μ1 ðtÞ represent the target gray value and the
1 + exp ð−avÞ
background gray value, respectively, and according to the
optimal gray threshold t obtained by the formula, it can be
Using the sigmoid function, the two operators of cross- obtained by formula (11):
over probability and mutation probability can be improved
to ensure that the curve is always in a smooth state at f.
d2 ðt Þ = ðu0 ðt Þ − u1 ðt ÞÞ2 : ð11Þ
3.2. Image Segmentation Based on Improved GA-IGA Image
Segmentation Method. When performing image segmenta-
tion, how to choose the most suitable grayscale threshold In order to obtain the optimal solution of d 2 ðtÞ, the opti-
for this segmentation in the grayscale image is the key to mization of d 2 ðtÞ can be realized by combining the evolution
effectively separating the target image from the background process of GA and this kind of GA is called improved GA
image. The grayscale threshold is to divide all the brightness IGA. The algorithm is realized by the combination of GA
values in the image into two categories higher than the and OTSU. A functional is a mapping from functions to
threshold and lower than the threshold according to the numbers. An operator is a mapping from functions to func-
specified brightness value (i.e., the threshold). If the back- tions. The improved IGA mainly realizes nonlinear adjust-
ground information points are mistakenly divided into tar- ment in the crossover and mutation operators. The
6 Applied Bionics and Biomechanics

(a) (b)

Figure 4: Human CT image segmentation. (a) Before scanning. (b) After scanning.

operator function is expressed as formula (12): then the algorithm uses the user’s input as the segmentation
constraints to automatically calculate the optimal segmenta-
8 p −p
> tion that meets the constraints. In this study, the liver tissue
>
<   cmax c min 
1 + exp A 2 f ′ − f /f max − f − 1 f ′ > f : segmentation of the initial slice image adopts the interactive
Pc = graph cut algorithm, and the label value vector of all pixels
>
>
: on the image is defined as Formula (14):
pc min , f ′ ≤ f
ð12Þ Àn n n Á
Ln0 = l1 0 , ⋯, li 0 , ⋯, lI 0 : ð14Þ
In this study, the classification error is used to evaluate
the pros and cons of the algorithm, which is usually ME ∈ ½ Since it is currently a binary classification problem, the
n
0, 1Š for the convenience of analysis. The calculation formula label value li 0 ∈ f0, 1g represents the label value of the ith
n
for misclassification is expressed as formula (13): pixel in the initial slice layer li 0 . The label value of 0 repre-
sents the background, otherwise it is the foreground. The
jG0 ∩ Gj + j F 0 ∩ F j segmentation energy function of the initial slice is defined
ME = 1 − : ð13Þ as formula (15):
jjG0 j + j F 0 jj
ÀnÁ  
n n
where G0 and F 0 represent the target and background EðLn0 Þ = 〠 R l1 0 + λ0 ∗ 〠 B li 0 , l j 0 : ð15Þ
areas in the original image when ideally segmented, i∈I ði, jÞ∈E

respectively.
Among them, pixels i and j are adjacent pixels connected
3.3. CT Image Segmentation Based on IGA Algorithm. If the by weights. λ0 is a constant coefficient, which measures the
input abdominal CT scan sequence traverses the cross- proportion of boundary items and area items.
sectional slice image sequence along the vertical axis from n
The area item Rðl1 0 Þ represents the penalty item with the
the top of the liver to the right lung lobe, the shape and area n0
label value l1 assigned to 0 and 1, respectively. If the gray
of the liver tissue area in the slice image would gradually value of the pixel i is gi , the formula of the area item is
become larger. This method is mainly based on the consis-
defined as formula (16):
tency and similarity of organs and tissues between CT image
sequences. The human CT image segmentation is shown in Àn Á À Á
Figure 4. R li 0 = 0 = −In P f g ðgi Þ : ð16Þ
Figure 4(a) is the segmentation of the CT image before
scanning, and Figure 4(b) is the segmentation of the CT Grayscale values refer to the color value of each object
image after scanning. The third layer of the whole sequence when converting a grayscale object to RGB. The relationship
was selected as the initial slice layer. After analysis, the liver between white and black is divided into several levels
tissue in this slice had a larger area, a clear edge, and a dis- according to the logarithmic relationship, which is called
tinct degree of distinction in the shape and outline of the “gray level.” When the pixel i is marked as 0, the region item
liver tissue. Only a small amount of initial manual labeling of the pixel is the negative logarithm of the gray value of the
is required to obtain accurate initial liver tissue segmentation pixel in the foreground gray histogram, and it is the opposite
contours. Interactive image segmentation means that the when it is marked as the foreground 1. The region item intu-
user first specifies part of the foreground and part of the itively reflects that the pixel i is marked as a penalty item and
background of the image by some interactive means, and they have corresponding labels.
Applied Bionics and Biomechanics 7

n n
The boundary Bðli 0 , l j 0 Þ indicates that adjacent pixels i Table 1: Thresholds and time-consuming conditions of the three
and j are penalized with different label values. The specific segmentation methods.
calculation formula is as shown in formula (17). Number of experiments 1 2 3 4 5
    1 Image threshold 115 115 115 115 115
n n n n
B li 0 , l j 0 = δ li 0 , l j 0 ×   : ð17Þ GA
Time (ms) 28.21 21.46 26.03 26.07 26.15
gi , g j + 1
Image threshold 130 130 130 130 130
OTSU
Time (ms) 13.37 14.28 14.52 14.43 14.55
Combining the above formulas to define the value of the Image threshold 132 132 132 132 132
label vector, the global optimal liver segmentation result of IGA
Time (ms) 10.02 9.15 9.53 9.87 9.66
the initial layer is obtained by constructing the minimization
solution formula, as shown in formula (18).
Table 2: Efficiency of three algorithms for segmenting edges.
L∗n0 = arg min EðLn0 Þ: ð18Þ
Number of experiments 1 2 3 4 5
Image threshold 70 70 70 70 70
The abdominal region of the human body is one of the GA
parts with the most tissues and organs, the most complex Time (%) 78 70 76 77 76
structure and a high incidence of disease. With the develop- Image threshold 85 85 85 85 85
OTSU
ment and popularization of computers in various industries, Time (%) 75 72 71 75 70
computer-aided disease diagnosis technology is becoming Image threshold 90 90 90 90 90
more and more mature. However, abdominal CT scan IGA
Time (%) 92 90 91 88 87
images are easily affected by the motion and noise of the tis-
sues affecting the body and the low resolution of the equip-
ment itself, which brings great challenges to the correct
segmentation of tissues and organs in CT scan images. The the OTSU method, and the IGA algorithm is more accurate.
improved image segmentation algorithm-IGA algorithm The comparison of the segmentation results shows that the
proposed in this study improves the segmentation accuracy IGA algorithm also has certain advantages in computing
and efficiency of traditional image segmentation algorithms. accuracy when segmenting complex image targets.
In order to compare the accuracy of the three algorithms
4. Experiment of CT Image Segmentation in the segmentation edge more accurately, the segmentation
experiment is continued. This study does not continue to
4.1. Experiments on the Segmentation Effect of Three Image segment the head target, but selects the human body contour
Segmentation Algorithms. All the algorithms in the experi- as the segmentation target after the image is grayed. On this
ment simulation of the algorithm in this study are completed basis, the accuracy of the head results after segmentation is
on the notebook, and the CPU and running memory are compared. The results are shown in Table 2.
2.4 GHz and 4 GB RAM, respectively. The algorithm imple- As shown in Table 2, since the image is segmented after
mentation software is MATLAB 2016a. In order to verify the grayscale, it can be found that the thresholds of the three
efficiency of the proposed algorithm, program model exper- algorithms are relatively stable, and the OTSU method is
iments are carried out on the OTSU segmentation algo- the most accurate but has poor real-time performance. GA
rithm, the GA algorithm, and the IGA algorithm, is still a short time, but the threshold changes are active,
respectively. In order to highlight the practicality, the images around 4 pixels. The threshold of IGA changes within 2
selected in this part are several unconstrained images pixels, but the time is about four times faster than that of
annotated. OTSU. It is more stable in both time and segmentation accu-
The experiment tests the calculation time of OTSU seg- racy. It is shown in Figure 5.
mentation algorithm, GA algorithm, and IGA algorithm, Figure 5(a) illustrates that the head’s contour is subject
and the number of experiments is five times. The thresholds to errors if the OTSU method is used to process the image
and time-consuming conditions of the three segmentation because the target contour is clearer but the interference is
methods are shown in Table 1. poor. The outcome of segmentation using the GA approach
As shown in Table 1, it can be seen that the IGA algo- is displayed in Figure 5(b). Although GA’s contour process-
rithm saves one-third of the time compared to the OTSU ing is less thorough than OTSU’s, the primary objective is
method, and is nearly 5 ms faster than the GA algorithm still evident. It displays the versatility and global search ben-
with the advantage of fast operation speed. In terms of the efits of the GA algorithm. The IGA algorithm can see that
accuracy of the threshold, although it is not as stable as the the human head target in the segmentation area is relatively
OTSU segmentation algorithm, the change of the threshold complete and the segmentation accuracy is relatively high.
is controlled within 2 pixels. It shows that IGA is more The three algorithms of GA, OTSU, and IGA are calcu-
real-time in the feedback of calculation results. According lated through 100 generations of evolution, and the adaptive
to human eye observation, the segmentation range of OTSU degrees are obtained as shown in Figure 6.
segmentation algorithm, GA algorithm, and IGA algorithm As shown in Figure 6, compared with GA, IGA and
is slightly different. The two GAs are slightly better than OTSU have a better adaptive degree in the evolution process,
8 Applied Bionics and Biomechanics

100 100

80 80
Percentage%

Percentage%
60 60

40 40

20 20

0 0
50 100 150 200 250 300 50 100 150 200 250 300

sample set sample set

OTSU method OTSU method
GA method GA method
IGA algorithm IGA algorithm
(a) (b)

Figure 5: Comparison of the effects of three methods for processing images. (a) Comparison of the effects of the three methods for
processing the contour of the image target. (b) Comparison of the effects of the three methods for eliminating interference.

the above curves, it can be seen that the adaptability and

robustness of IGA are better than GA and OTSU.
100 It is well known that usually if an algorithm uses double
adaptation, it may lead to contradictions in the internal
80 mechanism of the algorithm. In order to eliminate this con-
tradictory effect, a large number of parameter selection
experiments are hereby done. It is shown in Figure 7.
Percentage%

60 As can be seen from Figure 7(a), for different types of

test images, the proposed algorithm obviously has better seg-
mentation quality. The segmentation result has a higher
40 signal-to-noise ratio. Compared with other algorithms, the
segmented image has clearer target information, and the seg-
mented target image has a clearer outline. The segmented
20 outline information, for instance, is more evident than
other algorithms, as the graphic illustrates. Figure 7(b)
demonstrates that, despite of not appearing to be very obvi-
0
20 40 60 80 100
ous, a deeper examination of the segmentation outcomes
produced by the suggested algorithm is usually preferable.
number of iterations As a result, when compared to existing algorithms, the sug-
OTSU method IGA algorithm gested technique may yield excellent segmentation results
for a variety of image formats, demonstrating a wider range
GA method Optimal solution of applications.
Figure 6: Adaptability of the three methods.
4.2. Experiments on the Efficiency of the Three Algorithms.
Although image threshold segmentation has been studied
and the strong ability to get rid of them can also avoid local for many years and various algorithms emerge in an endless
precociousness. Only IGA can obtain the optimal solution in stream, the problem of image threshold segmentation
the 20th generation, although there is an oscillation diver- becomes more and more complicated due to the variety of
gence in the middle. But because of its strong adaptive abil- images and the sources of images. The existing algorithms
ity, it quickly returned to the normal computing state. From are basically specific algorithms that can only solve some
Applied Bionics and Biomechanics 9

100 100

80 80
Percentage%

Percentage%
60 60

40 40

20 20

0 0
20 40 60 80 100 20 40 60 80 100
number of iterations number of iterations
OTSU method OTSU method
GA method GA method
IGA algorithm IGA algorithm
(a) (b)

Figure 7: Segmentation effect of the three algorithms in the case of different image clarity. (a) The segmentation effect of the three
algorithms when the image is clear. (b) The segmentation effect of the three algorithms when the image is blurred.

Table 3: Classification errors after segmentation of medical images.

evaluation standard or an objective evaluation standard,
Images GA (%) OTSU (%) IGA (%) the algorithm proposed in this study has a good segmenta-
1 27 12 2
tion effect, especially in the processing of medical CT images
with relatively less complex details.
2 20 10 5
The method in this study is utilized to carry out seg-
3 22 16 4 mentation studies on group data to confirm the efficacy
4 38 18 8 of the method. The experimental data comes from lung
5 21 15 7 CT images of the First Affiliated Hospital of a university.
There are 10 groups of normal images, and the nodule size
is 4–30 mm. Each set of data contains 35–45 frames, and
Table 4: Medical image segmentation running time.
the analysis of the obtained segmentation results are
Images GA (s) OTSU (s) IGA (s) shown in Figure 8.
1 0.97 0.68 0.21
Figure 8(a)illustrates how far the OTSU method’s seg-
mentation effect differs from the manual method’s segmen-
2 0.90 0.75 0.26
tation effect. Figure 8(b) illustrates how the segmentation
3 0.93 0.72 0.23 impact of the GA approach is superior to that of
4 0.92 0.73 0.27 Figure 8(a). Manual segmentation still has a significantly
5 0.95 0.71 0.24 worse impact than automatic segmentation, though.
Figure 8(c) shows that the IGA method is more similar to
the results of manual segmentation and can obtain the com-
specific images, which is very limited. Tables 3 and 4 show plete lung parenchyma edge, avoiding the omission of lung
the classification errors of the three algorithms in the study nodules. The proposed method can effectively segment the
and the running time of the algorithms after segmenting lung parenchyma from normal and diseased lung images.
the five medical images in the experiment: The segmentation results obtained by applying the method
As shown in Tables 3 and 4, compared with other differ- in this study to the left and right lungs are very similar to
ent algorithms, the IGA algorithm proposed in this study the results obtained by the clinicians’ manual segmentation,
not only reduces the misclassification rate greatly. Moreover, and the obtained segmentation results are satisfactory to the
the overall running time is the shortest, and in most cases a clinicians.
satisfactory segmentation threshold can be achieved with Although this study chooses the OTSU method as the
fewer iteration steps. Therefore, whether it is an intuitive main algorithm for calculating the threshold based on theory
10 Applied Bionics and Biomechanics

300 300

250 250

200 200
number of pixels

number of pixels
150 150

100 100

50 50

0 0
2 4 6 8 10 2 4 6 8 10

number of images number of images

OTSUmethod GA algorithm
manual segmentation manual segmentation
(a) (b)
300

250

200
number of pixels

150

100

50

0
2 4 6 8 10

number of images

GIA algorithm
manual segmentation
(c)

Figure 8: Comparison of the proximity of three methods of lung CT image segmentation and manual segmentation. (a) The proximity
comparison between the segmentation effect of the OTSU method and the manual segmentation. (b) The proximity comparison between
the segmentation effect of the GA method and the manual segmentation. (c) The proximity comparison between the segmentation effect
of the IGA method and the manual segmentation.

and experiments, the variability of important links, such as 5. Conclusions

selection, crossover, and mutation during the operation of
the algorithm may lead to other algorithms participating Image segmentation is not only one of the main techniques
in the improvement. It is also an important direction for of image processing, but also one of the key techniques of
future research that the population has better self- object recognition and object detection. The main idea is
adaptability to ensure excellent robustness during global to separate the target information from the background
evolution. information. Then the target information is marked and
Applied Bionics and Biomechanics 11

identified through a certain algorithm. Finally, the target [8] Z. Yang, J. Lian, S. Li, Y. Guo, Y. Qi, and Y. Ma, “Heteroge-
information is extracted from the image. In this study, the neous SPCNN and its application in image segmentation,”
traditional image segmentation method has been improved, Neurocomputing, vol. 285, no. APR.12, pp. 196–203, 2018.
and an image segmentation method has been formed. In [9] E. Cuevas, H. Becerra, A. Luque, and M. A. Elaziz, “Fast multi-
order to improve the segmentation accuracy, the GA is feature image segmentation,” Applied Mathematical Model-
avoided to fall into a local optimum because of precocious- ling, vol. 90, no. 5, pp. 742–757, 2021.
ness. On the basis of improving the traditional GA and opti- [10] M. I. Alghamdi, “Neutrosophic set with adaptive neuro-fuzzy
mizing the single-dimensional OTSU method, this study inference system for liver tumor segmentation and classifica-
uses the different characteristics of the two algorithms to tion model,” International Journal of Neutrosophic Science,
form an improved algorithm IGA. Through experiments, it vol. 18, no. 2, pp. 174–185, 2022.
is found that the improved algorithm shows advantages in [11] P. F. Shan, “Image segmentation method based on K-mean
terms of calculation time, computational complexity, and algorithm,” EURASIP Journal on Image and Video Processing,
segmentation effect. It is especially suitable for segmentation vol. 2018, no. 1, 2018.
of CT images. Although the IGA proposed in this study has [12] Y. Zhao, H. Li, S. Wan et al., “Knowledge-aided convolutional
certain advantages over the algorithm before the improve- neural network for small organ segmentation,” IEEE Journal of
ment in the test results, in the experiment, the unscientific Biomedical and Health Informatics, vol. 23, no. 4, pp. 1363–
sample selection would cause obstacles to the operation. 1373, 2019.
The main research direction in the future is still to be attrib- [13] B. Shuai, Z. Zuo, B. Wang, and G. Wang, “Scene segmentation
uted to the optimization and selection of control parameters. with DAG-recurrent neural networks,” IEEE Transactions on
Pattern Analysis and Machine Intelligence, vol. 40, no. 6,
pp. 1480–1493, 2018.
Data Availability
[14] Y. Cheng, D. Wang, P. Zhou, and T. Zhang, “Model compres-
Data sharing not applicable to this article as no datasets were sion and acceleration for deep neural networks: the principles,
generated or analyzed during the current study. progress, and challenges,” IEEE Signal Processing Magazine,
vol. 35, no. 1, pp. 126–136, 2018.
Conflicts of Interest [15] X. M. Zhang and Q. L. Han, “State estimation for static neural
networks with time-varying delays based on an improved
The authors declare that they have no conflicts of interest.. reciprocally convex inequality,” IEEE Transactions on Neural
Networks & Learning Systems, vol. 29, no. 4, pp. 1376–1381,
2018.
References
[16] I. Aljarah, H. Faris, and S. Mirjalili, “Optimizing connection
[1] T. Suzuki, T. Shiotani, and M. Ohtsu, “Evaluation of cracking weights in neural networks using the whale optimization algo-
damage in freeze-thawed concrete using acoustic emission rithm,” Soft Computing, vol. 22, no. 1, pp. 1–15, 2018.
and X-ray CT image,” Construction & Building Materials, [17] P. Yin, S. Zhang, J. Lyu, S. Osher, Y. Qi, and J. Xin, “BinaryR-
vol. 136, pp. 619–626, 2017. elax: a relaxation approach for training deep neural networks
[2] S. Ha, M. Nguyen, and L. Patel, “Quantitative estimation of a with quantized weights,” SIAM Journal on Imaging Sciences,
ratio of intracranial cerebrospinal fluid volume to brain vol- vol. 11, no. 4, pp. 2205–2223, 2018.
ume based on segmentation of CT images in patients with [18] L. F. Wang, H. Wu, D. Y. Liu, D. Boutat, and Y. M. Chen,
extra-axial hematoma,” The Neuroradiology Journal, vol. 30, “Lur'e Postnikov Lyapunov functional technique to global
no. 1, pp. 10–14, 2017. Mittag-Leffler stability of fractional-order neural networks
[3] L. Dijk, C. L. Brouwer, A. J. Schaaf, Johannes G. M. Burgerhof, with piecewise constant argument,” Neurocomputing,
and R. J. H. M. Steenbakkers, “CT image biomarkers to vol. 302, pp. 23–32, 2018.
improve patient-specific prediction of radiation- induced [19] J. Cheng, P. S. Wang, and L. I. Gang, “Recent advances in effi-
xerostomia and sticky saliva,” Radiotherapy and Oncology, cient computation of deep convolutional neural networks,”
vol. 122, no. 2, pp. 185–191, 2017. Frontiers of Information Technology & Electronic Engineering,
[4] C. Zorn, E. Bauer, M.-L. Feffer et al., “Building and exploita- vol. 19, no. 1, pp. 64–77, 2018.
tion of learning curves to train radiographer students in X- [20] S. Mei, A. Montanari, and P. M. Nguyen, “A mean field view of
ray CT image postprocessing,” Journal of Medical Imaging the landscape of two-layer neural networks,” Proceedings of the
and Radiation Sciences, vol. 51, no. 1, pp. 173–181, 2020. National Academy of Sciences of the United States of America,
[5] X. Li, G. Feng, and J. Zhu, “An algorithm of l1 -norm and l0- vol. 115, no. 33, pp. E7665–E7671, 2018.
norm regularization algorithm for CT image reconstruction [21] H. He, F. Gang, and C. Jinde, “Robust state estimation for
from limited projection,” International Journal of Biomedical uncertain neural networks with time-varying delay,” Journal
Imaging, vol. 2020, no. 1, p. 6, 2020. of Jishou University (Natural Sciences Edition), vol. 19, no. 8,
[6] L. C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. pp. 1329–1339, 2019.
Yuille, “DeepLab: semantic image segmentation with deep
[22] E. Kaur, “Artificial intelligence techniques for cancer detection
convolutional nets, atrous convolution, and fully connected
in medical image processing: a review,” Turkish Journal of
CRFs,” IEEE Transactions on Pattern Analysis and Machine
Computer and Mathematics Education (TURCOMAT),
Intelligence, vol. 40, no. 4, pp. 834–848, 2018.
vol. 12, no. 2, pp. 2667–2673, 2021.
[7] R. Mariano, D. Oscar, M. Washington, and R. M. Alonso,
“Spatial sampling for image segmentation,” Computer Journal,
vol. 55, no. 3, pp. 313–324, 2018.