You are on page 1of 6

2021 IEEE 2nd International Conference on Information Technology,Big Data and Artificial Intelligence (ICIBA 2021)

An image segmentation method of lung


lymphatic tumors based on PET-CT images
2021 IEEE 2nd International Conference on Information Technology, Big Data and Artificial Intelligence (ICIBA) | 978-1-6654-2877-4/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICIBA52610.2021.9688245

Zilong Dong1,Tianchun Xu2,Yuan Kang1,Jing Lian1,3,Bin Shi2,3*


1. Lanzhou Jiaotong University, School of EleCTronic and Information Engineering, No. 88 West Anning Road,
Lanzhou, Gansu, P.R.China, 730070
2. Gansu Provincial Hospital, Medical Intelligence Laboratory, No. 204 Donggang West Road, Lanzhou, China, 730000
3. Lanzhou University, School of Information Science and Engineering, No. 222 South Tianshui Road, Lanzhou, Gansu,
P.R.China, 730000
3481823984@qq.com, 351360930@qq.com, 1598299527@qq.com, lian322scc@163.com
*Corresponding author:Bin Shi Email:13609330285@126.com

Abstract—To address the problem from lymphatic to the detection of lymphatic tumors. The appearance of p
tumor segmentation from PET-CT images, this paper et makes medical imaging technology reached a new
presents an image segmentation method based on pulse- height, and it is the most distinctive is a systemic
coupled neural network (PCNN). PET-CT image detection tomography, this is other instruments cannot do, in
can provide more detailed tumor information for later addition to the source lesion site, can also find systemic
diagnosis and treatment, while our adjustable fire- soft tissue parts and bone without metastatic lesions,
controlled MSPCNN model (AFC-MSPCNN) can segment which can facilitate the evaluation of treatment effect and
lung tumors effectively. Related experiment results show
guide the next treatment[3]. If the body tumor is compared
that our proposed method can improve the image
to the sea, then the pet is the beacon, the CT is the
segmentation accuracy than other competitive methods.
navigation map, people can accurately target.
Keywords—PET-CT; AFC-MSPCNN; image; segmentation II. RELATED WORK
I. INTRODUCTION A. Comparison test
As the increase of lung tumors is with the increase of PET-CT is through positron emission X light imaging
smoking people, it will be difficult to cure if the disease is technology, can achieve molecular level accurate imaging,
discovered later, so it is particularly important how the is one of the current means of tumor detection, although
disease can be detected earlier. At present, most of the the means belongs to radiation imaging, but the radiation
general population do not pay much attention to the time, impact for patients is minimal, combining pet and CT can
should be in the early stage, the tumor is difficult to be be comprehensive and detailed tumor discovery, and solve
checked by the conventional means, so the core of this as soon as possible.
design is to be able to check the patient's tumor quickly
PET-MR is to combine positron technology with
and accurately.
NMR resonance technology, combining the two to
In practical application, positron emission tomography achieve large functional metabolism also meets the
CT mainly uses digital image segmentation technology. molecular level imaging requirements. It can achieve
Through image processing and image segmentation to functional complementarity[4].
obtain the required data image, it can realize multiple
Between the two, PET-MR has great advantages over
analysis, diagnosis and later treatment planning for the
PET-CT[4], in soft tissue detection, and also in the
detection site, so as to realize the help of PET-CT image
detection of prostate and breast cancer.However, the main
to patients[1].To achieve this purpose, on the one hand, it is
purpose of this design is to solve the problem of lung
necessary to improve the accuracy of digital image
tumor. The two have little difference in the imaging
processing, improve the threshold of image segmentation
results, and the PET-MR instrument is more precious, the
and test the segmentation area; on the other hand, on the
detection cost is higher and less data.In conclusion, PET-
basis of Matlab programming, combine PET-CT image
CT was selected as the main instrument for this design..
and MATLAB code to achieve accurate image analysis,
and meet the image accuracy with computer programming B. Digital image processing
language as the core[2].
Digital image processing, also known as computer
This design is mainly through PET-CT image image processing, is to convert the image signal into a
expansion, both computed tomography graphics digital model through software, and also performs a series
technology.It is able to respond to diseased genes, of image processing (including: noise removal, enhancing
molecules, metabolites, etc., especially using positrons to signal intensity, dividing and extracting the target
label grapes 6 This function of sugar directly corresponds signal)[5].

978-1-6654-2877-4/21/$31.00 ©2021 IEEE 799


Authorized licensed use limited to: ULAKBIM UASL - KOCAELI UNIVERSITESI. Downloaded on May 19,2023 at 18:15:01 UTC from IEEE Xplore. Restrictions apply.
Image segmentation is to extract the meaningful feature III. PREPARE YOUR PAPER BEFORE STYLING
data of the image through the code, and make the
characteristic image identify and analyze the data content.
This image segmentation is AFC-MSPCNN, this
segmentation technology is based on SPCNN, combining
people.
The body vision system (HVS), able to provide more
accurate segmentation results, reduces operational
complexity. The calculation formula is as follows:

ij 1   NWijk1Y1d [ n - 1])
U ij [ n ]  e - U ij [ n - 1]  S(
1d

E ij [ n ]  e - E ij [ n - 1]  AVY ij [ n ]

1, if U ij [ n ]  E ij [ n ]
Y ij [ n ]  
 0, else

Among them, Wijk1A new weight matrix setting


method is used, so β=e because this model matrix is 1-2α,
the final expression is the following:

 0.1 0.15 0.1 


NW ij  e - 2   0.15 0 0.15 
 0.1 0.15 0.1 

Fig 1 Flow chart of the experimental operation


The parameters in this model adopt the setup method of
a FC-MSPCNN model where the P represents the preset As shown in the figure, the experiment mainly realized
number of iterations, as the total number of iterations of tumor detection and analysis by MATLAB software
an effective firing cycle neuron. Decay factor expression combined with PET-CT images. Firstly. the Matlab
in this model: program; second, MATLAB for pet image tumor analysis,
and finally, the results of image analysis.
S m in A. Read and display of the images
 =  ln
P Extract image data: using the function imread can
achieve any Matlab supported image file format in:
C. Expected results of the experiments MAP=imread (FILENAME, F M T') where: FILENME-is
(1) Read, write and display of the file format, mainly to the name of the read image file and FMT is the image
process the PET-CT images format[7].
(2) Basic operations of image processing, providing Read the image information: use the function imfinfo to
linear operations such as sum and difference, as well as obtain the information about the image file, in the
nonlinear calculations such as convolution, correlation following format: INF0=imfinfofFILENAME, 'FMT'),
and filtering. INFO is the structure of the PET-CT image.
(3) Image transformation, and the PET-CT images are
converted into a common jpg format through a discrete Display image results: use the function imshow as the
display function in the following format: imshow (1, [ow,
Fourier transformation, facilitating follow-up operations
and contrast algorithms[6]. high])
(4) Image analysis and enhancement provide correction,
B. mainly contrasts the threshold segmentation method
histogram balance, median filtering, contrast adjustment
and other processing. Select Otsu threshold selection method, image
(5) Mathematical morphological processing with segmentation, with simple and fast processing
MATLAB as the core tool. characteristics[8], is a common valve value selection
method, mainly selects the Praytiosh function provided in
the MATLAB toolbox as the que value calculation
method.Its basic ideas are:

800
Authorized licensed use limited to: ULAKBIM UASL - KOCAELI UNIVERSITESI. Downloaded on May 19,2023 at 18:15:01 UTC from IEEE Xplore. Restrictions apply.
(1) The image prime was set to N, gray scale i to ni, IV. EXPERIMENTAL RESULTS
The probability is Pi=ni/N i=0,1,2,3,.....,M-1; A. raw image analysis
(2) Image grayscale values were divided into C by the
threshold N0And C1, Among them C0The pixel interval
of is [0.T],C1The interval of is [T+ 1, M-1], available: A B
T t  M -1
μ0 = 
t0
ip i /  0 μ1 = 
t  T 1
ip i /  1
T t  M -1
0 = 
t0
pi 1 =  pi  1-0
t  T 1
C D
(3)  2  0(1 0  1)
2
Combining the above variance
numbers, select the best threshold for the maximum value
of Otsu in the region range. Fig. 2 PET-CT original image. Image (a) original CT image; Image (b)
preprocessing segmentation result; Image (c) thermal imaging result
C. image segmentation technology from human body contrast map.
The methods used in this technique are: image edge
detection; region growth; region separation and As shown in the figure, the above image is based on the
aggregation, etc.The calculation is usually based on PET-CT image, and the human lung tumor image
similarity and discontinuity, which mainly refers to the obtained through Matlab processing. By analyzing the
segmentation of the image into similar regions according images, the lower lobe of the right lung showed irregular
to prior formulated guidelines[9], and the discontinuity is mass density, lobation, about 63 × 34 × 28mm, with
mainly based on discontinuous changes in the image gray heterogeneous abnormal radio-profusion, proximal
scale. vascular shadow of SUVmax8.90, lesion, bronchial
truncated, and plaque, cord and flocculation high density
Edge detection, as the main method, mainly can greatly shadows scattered around the distal end. Irregular mass in
reduce the amount of data, and can remove irrelevant the right lower lobe and abnormal FDG uptake. Combined
information, retain the important structural attributes of with biopsy pathology, consider lung malignancy,
the image, wherein the edge of the image refers to the area enlarged lymph nodes in zone 7 and right lung gate,
where the pixel gray scale changes step around the image, abnormally increased FDG uptake, highly suspicious
that is, where the gray scale value derivative is large. metastasis; right lower pulmonary obstructive pneumonia,
Basic steps of edge detection: smoothing filtering, systemic prostate enlargement; bilateral testicular sheath
sharpening filtering, edge determination, and edge effusion; and right lower hip calcification.
connection[10].
B. Comparative analysis of
D. The eigenvalue extraction is the outpu
The images with small differences in similar images
were finally determined, and the data with the largest
eigenvalues were extracted by selection judgment. This
method is performed in three steps, first by a simple
mapping of its edges (including perimeter, area, density,
size mean and Euler number), followed by the main
component fraction [COEFF,SCORE,1atent]=pincomp
(X).

Fig 3 Image processing process

As shown, the contrast plots obtained by MATLAB


code runs of the PET-CT images were separated, followed
by the image segmentation to separate the tumor region.

801
Authorized licensed use limited to: ULAKBIM UASL - KOCAELI UNIVERSITESI. Downloaded on May 19,2023 at 18:15:01 UTC from IEEE Xplore. Restrictions apply.
C. Comparative analysis

(a)AFC-MSPCNN (b)SPCNN (c)OTSU (d)KITTLER (e)Pre'treatment (f)PET-CT


Fig 4 PET-CT original image

As shown in Fig, (a) is the segmentation method used accuracy, which is because I used the expansion algorithm
for this design, (b) (c) (d) is the contrast experiment, (e) is to remove other influencing factors and changed the focal
the result of preprocessing, and (f) is the PET-CT image area, resulting in some error in the results.
of this selection.This image preprocessing mainly adopts
expansion operation, this algorithm can be able to
improve the brightness of tumor area, increase tumor area
to reduce other areas, compare figure (a) (b) (c) (d), we
can find that AFC-MSPCNN still has obvious advantages,
it can provide more smooth and accurate tumor image,
which can also greatly improve the accuracy of the
experiment, followed by OTSU method also by strong
contrast[11],A good comparison in the later
experiment.However, when the doctor directly observes
the image, there may be some error in the result
estimation, so we chose a set of rating criteria to analyze
the images.As shown in the table below, by comparing
precision and Recall values, we can find that AFC-
MSPCNN processes other methods in both accuracy and
image integrity, but it is difficult to reach 100% in

802
Authorized licensed use limited to: ULAKBIM UASL - KOCAELI UNIVERSITESI. Downloaded on May 19,2023 at 18:15:01 UTC from IEEE Xplore. Restrictions apply.
Table I. Evaluation criteria

image algorithm Dice VOE precision Recall

AFC-MSPCNN 0.9895 0.0144 0.9824 0.9966

SPCNN 0.9615 0.0145 0.9644 0.9888

PET-CT-1 OTSU 0.9916 0.0161 0.9437 0.9797

KITT L ER 0.8911 0.0173 0.9126 0.9197

Pre'treatment 0.7367 0.2263 0.8611 0.7237

AFC-MSPCNN 0.7890 0.4212 0.6517 0.9985

SPCNN 0.7876 0.4243 0.6498 0.9998

PET-CT-2 OTSU 0.7871 0.4253 0.6490 0.9996

KITT L ER 0.7865 0.4266 0.6482 0.9998

Pre'treatment 0.7375 0.4945 0.5667 0.8743

AFC-MSPCNN 0.7659 0.2771 0.7959 0.6754

SPCNN 0.7106 0.3749 0.7811 0.4456

PET-CT-3 OTSU 0.6793 0.3578 0.8492 0.5645

KITT L ER 0.5348 0.4893 0.6271 0.5907

Pre'treatment 0.4547 0.4432 0.5179 0.3423

AFC-MSPCNN 0.9642 0.1563 0.9906 0.8654


SPCNN 0.9193 0.2201 0.9760 0.7734

PET-CT-4 OTSU 0.8945 0.2638 0.9425 0.6989

KITT L ER 0.7997 0.4274 0.8717 0.7021

Pre'treatment 0.7651 0.3437 0.7860 0.5784

AFC-MSPCNN 0.8771 0.3244 0.8892 0.8857

SPCNN 0.8114 0.3421 0.8653 0.8504

PET-CT-5 OTSU 0.7971 0.3678 0.7865 0.7988

KITT L ER 0.5412 0.3743 0.5589 0.8896

Pre'treatment 0.7000 0.3645 0.6783 0.6345

V. CONCLUSIONS simple and effective.There are still some deficiencies in


this experiment. Due to the choice of expansion operation,
This paper is mainly designed to realize the analysis there may be some errors in the analysis of the lesion, and
of PET-CT images of lung tumors through Matlab. the solution combines expansion and corrosion to
Traditional manual analysis of CT images is easy to cause constitute an open or closed operation.In conclusion, we
error in the analysis results, which is not conducive to can find that this experiment is relatively successful, both
later treatment, and Matlab-based image processing can in image processing or in result analysis, and has good
well avoid the above problems.After using methods such reference value in future tumor detection.
as the A FC-MSPCNN technology segmentation
algorithm and edge detection, we can clearly observe that ACKNOWLEDGMENT
it not only reduces the computational complexity, but also
The authors acknowledge the support and assistance
improves the smoothness and clarity of the
provided by Bin Shi and Jing Lian in my work, as well as
image.Combined with practical application, this method
the encouragement of my work by colleagues at Yuan
can be determined as an effective medical image
Kang. This study was funded in National Natural Science
processing technology, making tumor detection more
Foundation of China (grant numbers 62061023 &

803
Authorized licensed use limited to: ULAKBIM UASL - KOCAELI UNIVERSITESI. Downloaded on May 19,2023 at 18:15:01 UTC from IEEE Xplore. Restrictions apply.
61941109 & 61861024 ), Natural Science Foundation of Conference: 2021 IEEE 4th Advanced Information Management,
Gansu Province of China (grant numbers 21JR1RA024 & Communicates, Electronic andAutomation Control Conference
(IMCEC), 2021, (01):1-7.
21JR1RA252) and Science and Technology Plan of
[7] M. Kostas, The Constantly Evolving Role of Medical Image
Gansu Province(grant number 20JR10RA273). Processing in Oncology: From Traditional Medical Image
Processing to Imaging Biomarkers and Radiomics [J], Journal of
REFERENCES imaging, 2021, 7(8): 124-124.
[1] Yan Jie, Li Xiaolin, and Xu Hui, Application of F-FDG PET/CT [8] Chai Guangjin, Li Weiwei, Lu Bo, Xu Man.PET/CT metabolic
imaging combined with tumor markers in cervical cancer parameters with LMR in recent clinical studies assessing the
diagnosis [J].China CT and MRI Magazine, 2021, 19(01): 155- recent efficacy of chemoradiotherapy in lung cancer [J].China CT
157. and MRI Magazine, 2021,19 (08): 54-56.
[2] Zhang Chong, Diagnosis value of magnetic resonance class PET [9] F Janata, N Fezoulidis, O Barachini, S Mirzaei, K Hergan, S
imaging technology for malignant tumor metastasis [J]. Zandieh. Common hepatic artery aneurysm detected by 18F-FDG
Psychological Monthly, 2020, 15(13): 140. PET/CT Imaging [J]. Radiology Case Reports, Volume 16, Issue
[3] Lian Jing, Yang Zhen, Sun Wenhao, Guo Yanan, Zheng Li, Li 11. 2021, PP 3157-3161.
Jinping, Shi Bin, and Ma Yide, An image segmentation method of [10] [10] Gao Xinle, Dong Liuyi, Wang Chong, Liu Meiqin, Wei Lei,
a modified SPCNN based on human visual system in medical Xu Peng, Liu Yan, Hu Zongtao.Clinical significance of systemic
images [J]. Neurocomputing, 2019, 333: 292-306. PET/CT tomography combined with tumor markers in the
[4] F. Sampedro, S. Escalera, A. Domenech, and I. Carrio, A diagnosis and differential diagnosis of lung cancer [J].Anhui
computational framework for cancer response assessment based Medicine, 2015,19 (04): 702-705.
on oncological PET-CT scans [J]. Computers in Biology and [11] Ding Wenxiang, Yu Jiangyuan, Zheng Chaojie, Fu Peng, Huang
Medicine, 2014, 55: 92-99. Qiu, F. David Dagan, Yang Zhi, Wahl Richard L, Zhou Yun,
[5] Lian Jing, Yang Zhen, Sun Wenhao, Zheng Li, Ma Yide, Qi Machine learning-based noninvasive quantification of single-
Yunliang, A fire-controlled MSPCNN and its applications for imaging session dual-tracer 18F-FDG and 68Ga-DOTATATE
image processing [J]. Neurocomputing, 2021, 422: 150-164. dynamic PET-CT in oncology [J]. IEEE transactions on medical
imaging, 2021.
[6] Huang Ruifeng. Lian Jing. Zhang Caixia, Multi-step breast mass
image segmentation method based on MFC-PCNN [C].

804
Authorized licensed use limited to: ULAKBIM UASL - KOCAELI UNIVERSITESI. Downloaded on May 19,2023 at 18:15:01 UTC from IEEE Xplore. Restrictions apply.

You might also like