View of Automatic Segmentation of Lung Tumors Using Adaptive Neuron-Fuzzy Inference System

(1)

17469 http://annalsofrscb.ro

Automatic Segmentation of Lung Tumors Using Adaptive Neuron-Fuzzy Inference System

*1S.Dhanasekaran,²Dr.P.Mathiyalagan,³Rajeshwaran, ⁴A.Manikandan

1Assistant Professor, Department of ECE, Sri Eshwar College of Engineering, Coimbatore

2Associate Professor, Dept. of CSE, Sri Ramakrishna Engineering College, Coimbatore

3Assistant Professor, Department of ECE, Sri Ramakrishna Engineering College, Coimbatore

4Assistant Professor, Department of ECE, Vivekanandha College of Technology for Women, Namakkal

1[email protected], ²[email protected], ³[email protected] ,

4[email protected]

ABSTRACT

Cancer is one of the deadliest diseases in the world, and it is still on the rise. More than 200 types of cancer have spread to humans. Over the past few decades, lung cancer (also called lung cancer) has been at a high level among all cancers. CT images are the most commonly approach used for Lung cancer treatment. Improved isolation results can provide better diagnostic results and help surgeons treat correctly. According to recent research, image pixel ratio calculation is used in image analysis to detect the characteristics of CT images. It is still useful to use an unbiased metaphor to classify and identify abnormal problems, especially for various cancers such as lung cancer. In this research, we proposed a FBB (Fast Bounding Box) method, which is an advanced segmentation algorithm that uses the gene segmentation algorithm used in the CT image separation process. The genetic algorithm is used to calculate the threshold, and the noise -reduced image is used to detect the number of movements and tumors. The GLCM2 algorithm is used to select segmented image features. The Adaptive Neuron Clearing Inference System (ANFIS) is used in this type of lung, which is in the normal, intermediate or abnormal stage. Comparing the results with the results of the river basin division, it shows a significant improvement.

Keywords: Lung Cancer, Lung CT scan image, Image Segmentation, Feature Extraction, Adaptive neuron-Fuzzy Inference System.

I. INTRODUCTION

The lung cancer is the prominent reason for cancer death. Tobacco Smoking is considered to be the prime factor for Lung cancer in both male and female. But there are some cases the Non-smokers also get affected by Lung cancer, mainly due to exposure to toxic gases and passive smoking. The WHO (World Health Organization) presents 2015 world health statistics on lung cancer which figures out an estimated 157,499 people death rate [1-3].

There are major contributions from Medical, Biological, Computer, Electronics and Electrical fields to annihilate Lung cancer disease from mankind. Even today the success rate for cancer treatment is very less [4].

The lungs are pair of mushy air-filled parts situated on both sides of the chest. The trachea allows breathe in air shooting up to the lungs clinched the cylindrical burning

(2)

termed bronchi [6]. The bronchi are divided into smaller bronchioles and finally becoming microscopic. The bronchioles lastly terminated with collections of minute air sacs which are known as beamed alveoli. In the alveoli, oxygen in the air is absorbed into blood. Resonant layers of watery substance are acting as a lubricant which allows lungs to glide easily as they develop in control to pact through each breathes [7].

Depending on the properties of the infected cells, there exist two categories of cancer.

They stay kindly and malicious forms [8]. The non-threatening cell is inadequate to self- reliant quarantines in the body that do not entrench or overdo to original positions. In contrast, malignant cancer cells are represented by an indiscriminate evolutionary progression, disrupting and terminating in nearby cells, and spreading to other areas on the side of the lung [9-11].

The assessment of the lung tumor of patients is very informative in clinical practice. The most important mission is found as the initial recognition of tumor in lungs and to decide on appropriate treatments to protect lungs [12]. Lung CT scan images are been taken for image processing steps and in which segmentation phase is a pivotal and necessary constituent of investigation on duplicate classification [13]. The accurateness and difficult assessment confidence value of every lung abnormality detection system based on an effective lung segmentation method [14]. The CT image is been subdivided in to many divisions called segments. The effect of image splitting up is an accumulation of slices which syndicate to arrangement the all-inclusive image. There are different methods have been proposed intended for better and improvised segmentation process [15].

II RELATED WORK

The bounding box segmentation within axial brain MR images. This is completely unsupervised and also very efficient novel segmentation [16]. The Otsu’s thresholding Systems designed for the integration of cancerous or tumor region from CT Medical Images.

It remains a stretch overriding progression which finds high noise in segmented image [17].

From the author suggested that amount of cancer level of male and female in 2010 that report will be consider here with higher level based on the all other guide advices.

The Fast Bounding algorithm employs the in innovative total occupation that can ascertain the region of conversion by means of brain image vertically and horizontally [18].

Very importantly the phase in separating these methods is to understand and analysis of co-occurrences, primary to the identification of lung cancer sub-types. Now, by analyzing the changes between the malignant cells and the cells that have formed and the original normal cells, modeling them as indistinguishable, and modeling them as beneficial changes [19].

This previously has intense implications for the clinical supervision of patients, guidelines of cancer medicine – the traditional classifications of malevolence.

(3)

Image segmentation is an important field in the field of image processing, an important step in the image processing technique, and this is the focus [20]. Image segmentation can’t be put under one category and there is no common solution arrived for image segmentation.

These algorithms frequently have to be united with specific domain knowledge in order to achieve efficient image segmentation [21].

The lung isolation algorithm can accurately segment lung tumors on lung CT scans. The algorithm was tested on a data set of 25 different patients [22]. The gray-level thresholding near sector the lungs surrounded by every single subtracted tomography subdivision [23]. The analysis concluded only inaccurate results along the forward join line, riddance of the trachea and bronchi from the lung integration areas, besides of elimination of the diaphragm regions.

II. METHODS AND MATERIAL

The Lung CT image is acquired in digital form and it can’t be used as it is, this is because of noise and distortions that present in the image. Image enhancement is also required to improvise the quality of the input image. Here, in this method will support different image types such as JPEG, PNG.

A. Preprocessing

The input image is changed to 256x256 and converted to a binary image with black or white pixels only for better performance. Mainly, the objective of applying enhancement techniques is to fetch out features that are concealed, or just to focus on definite features of input image such as, altering brightness, contrast etc. [7]

The gray scale image is put in to gradient process on x and y directions to implement directional variations in intensity as well as color of the input image. Smoothing is also performed to gain important patterns available in the image and eliminating noise values from the image.

Disorder is the unwanted distinctions or dissimilarity of the image. This process is for smoothing the duplicate by clean the edges and structures. Consistent states are vastly flattened and solid edge constituencies are located hardly smoothed. It is used because of high precision and dependability.

B. Segmentation

Image segmentation is a process to sustain the image properties in the perfect tumour detection and classification. Image segmentation process makes the image processing easier.

It is the method to split or subdivide the image as segments [1]. The main objective of this research is to propose a fast and efficient segmentation algorithm, which detect lung nodules more accurately.

(4)

Genetic Algorithm is used for segmentation process. Here, image is been denoised using Gaussian filtering method. Genetic Algorithm uses fitness function to calculate the size of the population and also it calculates Lung image edge detection and restoration values.

The subdivision is the course of splitting a numerical appearance obsessed by several slices. The main function about the lung disease would control by the cancer rays analysis (ranks, arcs, etc.). The end result of double separation is arrangement of sectors that protections the total copy or a traditional of reliefs extracted from the identical. Every single pixel in the section is matching with respect to some characteristics or considered possessions, such as color, dissimilarity, intensity, or texture.

C. Segmentation using Fast Boundary Box (FBB)

FBB [2] is a new fast sub division method that habits balance to encompass an incongruity (typically tumors) by bounding box within a CT lung images. In each input CT slice; there is a left–right federation of equilibrium of the lung.

[8] Vertical direction of the image Bhattacharya coefficient (BC) measures the resemblance between normalized gray close force histograms of horizontal and vertical direction when its value 1 is normalized and remains 0 when the histograms differ. Regions, then Bhattacharya coefficient value position is separated from the healthy textures.

Adaptive Neuron-Fuzzy Inference System (ANFIS)

This research paper obligate only painstaking affiliation functions that have stood steady, and rather arbitrarily elected as shown in Figure 1.

Figure 1: Adaptive Neuron Fuzzy Inference System

• Also, we force only functional fuzzy corollary to displaying systems whose rule contraction is to all targets and resolve the characteristics of the variables in the perfect regions.

• In general the outline of the affiliation occupations rest on parameters that can be adjusted to membership functions.

• The parameters can be regularly adjusted tentatively on the data that we try to model.

In this paper, the proposed segmentation algorithm for tumor segmentation of CT lung image involves two steps

(5)

1. The paramount point encompasses amputation of unwanted parts.

2. The second stage is to remove noise and piecewise smoothening. The edges are preserved for further segmentation process. The anisotropic transmission strainer is recycled on behalf of better contrast and noise removal as shown in Figure 2.

Figure 2: Flow chart of proposed method

Data Mining Preprocessing

Data Mining is also clubbed with image processing to achieve better performance in the desired output image. Data Mining is the process to dig up the unknown patterns by comparing input image patterns with predefined training data sets. At this stage, the image is essential to evaluate the authority of the analyses and makes the feature mining phase more reliable as shown in Figure 3. Data mining is considered as the part of lung cancer cure, Data Change, Data cleaning and Data Configuring. Data mining process includes training coupled dictionaries for super-selection.

Figure 3: Data mining

The proposed algorithm uses fast boundary box segmentation which is unsupervised change detection method to detect the region of interest.

(6)

Algorithm:

1. Input Lung CT scans image acquisition.

2. Perform Edging & Smoothing for the input image 3. Calculate region based detected image.

4. Input image is trained with training images (data sets) 5. Find edges and apply filter to denoise.

6. Apply Genetic algorithm

7. Use fitness function value abs (-(1/4)*z^2 +2*z +5)

8. Generates the fitness from Fitness values using Fitness function.

9. Threshold calculation using genetic algorithm 10. Genetic algorithm based edge restoration.

11. Detect Region Segmented Outlines.

12. Apply Fast Boundary Box Processing steps.

13. Genetic Algorithm Final Training Count and Tumor Detection Processing 14. Genetic Algorithm Feature Selection Processing.

15. Find image feature values.

16. Apply ANN Classification

17. Calculate Performance Accuracy values.

(7)

III. Existing Method

The input CT Lung image is obtained and been implemented in watershed segmentation algorithm which works based on topological gradient model [14] as shown in Figure 4.

Figure 4: Lung CT Input Image

The anisotropic diffusion filter image hazes over areas of image that is small gradient magnitude. This method is the one of the efficient image enhancer and also restorer based on Partial Differential Equations.

Filter image

• Most of the input images are been affected with some noise that is unexplained variations or disturbances of data. Image analysis becomes effective when these noises are been filtered out using filters. The existing method uses median filters to eliminate the noises and produces filtered image. Filters are generally used to enhance or remove certain image properties. The filtering method includes sharpening, smoothing the image and edge enhancement.

• The outcome is to discard solid outlier low as well as high pixel intensities although having a comparatively trivial consequence on the image specifics compared to other smoothing techniques. This filter is mostly suitable for eliminating cosmic ray points from images that have comparatively large images as shown in Figure 5

Figure 5: Filter Image

(8)

Binary image

The input image will be converted in to binary image. The boundaries in binary image that subtract the binary conditional categories. Apply pressure to muscles to give freeness in the tissue while enter into the binary scale section gray scale image to binary image. From this basic simulation shows how the binary section will work to determine the lungs cancer disorders. The threshold value of the image is measured using standard deviation values and the binary image is been created as shown in Figure 6.

Figure 6: Binary Image Segmentation image

The conventional watershed method used and it eliminates components of lesser specified pixel size from the binary image and produces resultant binary image. This process is called as an area opening. Image region properties and label connected component values are implemented to generate segmented image as shown in Figure 7.

Figure 7: Segmentation Image

(9)

A region segment outlines lung cancer detection

The main function about region outlines detection having better simplicity, low noise and fiction. The lung tumor portions are outlined using region boundary values based on region property values as shown in the Figure 8.

Figure 8: Region Segment Outlines Lung Cancer Detection

Genetic algorithm is been implemented to count and detect the number of regions based on pixel property values as shown in Figure 9. With this input image there are four regions are detected and numbered.

Figure 9: Genetic Algorithm Lung Cancer Count and Detection

GLCM FEATURE SELECTION PROCESSING

Gray-Level Co-Occurrence Matrix (GLCM) is an m x n x p array of valid gray-level co- occurrence matrices is been implemented for Lung CT image feature selection processing [10]. The segmented image GLCM properties are listed below:

Sensitivity = 3.3675 Specificity = 90.0000 Accuracy = 95.0000 Contrast = 0.0001 Correlation =0.0022 Energy =0.8595 Homogeneity =0.9822 Standard deviation =0.9989

(10)

Mean = 0.0881

IV. PROPOSED METHOD RESULT

Input Image

In this proposed method, the same Lung CT image is taken for further image processing and analysis. This proposed system is designed to support various image types such as JPEG, PNG, and BMP etc as shown in Figure 10.

Figure 10: Lung CT Input Image

Filter Image

The technology is used to identify accurate edges of the input gray scale image, typically boundaries, lines or other intricacies that stand authoritative projected for the understanding of the duplicate segments. Gradient images with X and Y directions, edges are been calculated and smoothing and region based images are created. Canny edge detection method is been implemented to produce better results as shown in Figure 11.

Figure 11: Filter Image

Denoised image

The image at this stage of image processing is denoised. That is, delete the unexplained data already in the image, and define the border and bright area, as shown in the Figure 12.

(11)

Figure 12: Denoised Images

Proposed Segmentation image

Image segmentation is done using the proposed algorithm, using Genetic Algorithm. Initially, the image is been subdivided in to patches of size 10. The input images are been trained with set of training images. Cutting the two main spots from the gray scale image and eradicate the annoying blob to get two primary counties.

Figure 13: Segment Image

Figure 14: Segmented Image and Detection of Lung Cancer

(12)

Table 2: Proposed Method Image Feature Values

Region number

Intensity Area perimeter Centroid Diameter

1

327.0 99.5 180.3 132.2 20.4

In the Lung CT image, the detected tumors are detected and rectangle shape boundaries are marked by implementing boundary box and region properties as given Table 2. The center image of the Figure 13 and 14 shows the results for boundary box segmentation method.

Genetic Algorithm Feature Selection Processing is been implemented to count the tumor affected regions. The regions are numbed and intensity, area, perimeter, centroid values, and diameter of each lung tumor regions are measured.

Conclusion:

In this article an efficient algorithm for lung tumor segmentation is proposed. Tumor images are segmented by using fast bounding box segmentation algorithm. The results shows the tumors are of different dimensions and different stages accurately. The proposed segmentation algorithm measures intensity, area, perimeter, centroid and diameter values of each tumor region accurately. The tumor affected boundary area is perfectly outlined with boundary box method and also tumor borders are indicated with accurate outlines.

Since, boundary box method is combined with genetic algorithm features, more accurate results are achieved than the existing segmentation algorithm. The proposed algorithm took very lesser time for entire segmentation process when compared to the existing system.

The segmented image features such as PSNR, MSE, sensitivity, specificity, accuracy, processing time, contrast, correlation, energy, homogeneity, standard deviation and mean values are compared and the proposed algorithm shows better improvements.

PSNR and MSE values are calculated and compared on both existing and proposed method; the result shows that the proposed method gives high PSNR and low MSE values. It shows that the working of proposed algorithm to detection in lung tumor is better.

In future, breathing rejection measurement, vector, clustered vector, timing accuracy values can be added as parameter values in addition to intensity, area, iteration, perimeter, centroid and diameter values to get the better and perfect simulation output. This method may fine-tune to detect the time of berating and lungs function of patients. Also, give recommendations about treatment procedures, surgery, radiation therapy, chemotherapy, immunotherapy, Targeted therapy and hormone therapy and suggest medicines to the physicians based on the tumor classification.

(13)

REFERENCES

[1]. Bhagwati Charan Patel, Sinha G. R.(2010), ‘An adaptive K-means clustering algorithm for breast image segmentation’, International Journal of Computer Applications,Vol.

10,,No. 4, pp. 35-38.

[2]. Baidya Nata Saha, Nilajan Ray, Russell Greiner, Albert Murtha, Hong Zhang (2012),

‘Quick detection of brain tumors and edemas: A bounding box method using symmetry’, Computerized Medical Imaging and Graphics 36, pp 95-107.

[3]. Faleh H. Mahmood, Wafaa A. Abbas, Ali .S.M. (2014), ‘Estimating lung tumor size in CT image using image segmentation technique’, International Journal of Emerging Technology and Advanced Engineering (IJETAE) – Vol 8 No. 3.

[4]. Manikandan, A., & Sakthivel, J. (2017a). Recognizable Proof of Biometric System With Even Distorted And Rectification States. Journal of Advanced Research in Dynamical and Control Systems, 9(2), 1393–1398

[5]. Sarthak panda (2015), ‘Color Image Segmentation Using K-means Clustering and Thresholding Technique’, International Journal of Engineering Science and Computing, Research article. May 2015.

[6]. Fatma Taher, Naoufel Werghi and Hussain Al-Ahmad (2015), ‘Computer Aided Diagnosis System for Early Lung Cancer Detection’, Algorithms, Vol. 8, PP 1088- 1110

[7]. Khin Mya Mya Tun, Aung SoeKhaing (2014) ‘Feature Extraction and Classification of Lung Cancer Nodule using Image Processing Technique’, International Journal of Engineering Research and Technology (IJERT), ISSN: 2278-0181 Vol.3, Issue 3.

[8]. Rohan Kandwal and Ashok Kumar (2014), ‘An Automated System for Brain Tumor Detection and Segmentation’, International Journal of Advanced Research in Computer Science and Software Engineering, Vol. 4, Issue 3.

[9]. Saeid Fazli and Parisa Nadirkhanlou (2015) ‘A Novel method for Automatic Segmentation of brain tumor in MRI images’, International Journal of computer vision and pattern recognition, Vol.3, No.2.

[10]. Shahid Eqbal, Ansari M.A (2015), ‘Medical Image Feature Extraction for Computer Aided Diagnosis of Lung Cancer’, International Journal of Advanced Research in Computer Science and Software Engineering, Vol.5, Issue 6.

[11]. Rakesh M,Ravi T(2012),“Image Segmentation and Detection of Tumor Objects in MR Brain Images using fuzzy c-means (FCM) Algorithm”, International Journal of Engineering Research and Applications,Vol.2,Issue3,pp.2088-2094.

[12]. Anter, Ahmed M., and Aboul Ella Hassenian. "Computational intelligence optimization approach based on particle swarm optimizer and neutrosophic set for abdominal CT liver tumor segmentation." Journal of Computational Science 25 (2018): 376-387.

[13]. Manikandan, A., & Jamuna, V. (2017). Single Image Super Resolution via FRI Reconstruction Method. Journal of Advanced Research in Dynamical and Control Systems, 9(2), 23–28

[14]. Petrick, Nicholas, BerkmanSahiner, Samuel G. Armato III, Alberto Bert, LoredanaCorreale, Silvia Delsanto, Matthew T. Freedman et al. "Evaluation of

(14)

computer‐aided detection and diagnosis systems a." Medical physics 40, no. 8 (2013):

087001.

[15]. Henriksen, Emilie L., Jonathan F. Carlsen, Ilse MM Vejborg, Michael B. Nielsen, and Carsten A. Lauridsen. "The efficacy of using computer-aided detection (CAD) for detection of breast cancer in mammography screening: a systematic review." ActaRadiologica 60, no. 1 (2019): 13-18.

[16]. Manikandan, T., N. Bharathi, M. Sathish, and V. Asokan. "Hybrid Neuro-Fuzzy System for Prediction of Lung Diseases Based on the Observed Symptom Values." Journal of Chemical and Pharmaceutical Sciences ISSN 974 (2017): 2115.

[17]. Malviya, Neha, Naveen Choudhary, and Kalpana Jain. "Content based medical image retrieval and clustering based segmentation to diagnose lung cancer." Advances in Computational Sciences and Technology 10, no. 6 (2017): 1577-1594.

[18]. Wason, JeyaprakashVasanth, and AyyappanNagarajan. "Image processing techniques for analyzing CT scan images towards the early detection of lung cancer." Bioinformation 15, no. 8 (2019): 596-599.

[19]. Khalid, Shehzad, and Uzma Jamil. "Valuable pre-processing and segmentation techniques used in automated skin lesion detection systems." (2015).

[20]. Chaddad, Ahmad. "Automated feature extraction in brain tumor by magnetic resonance imaging using gaussian mixture models." Journal of Biomedical Imaging 2015 (2015): 8.

[21]. Manikandan, A., Suganya, K., Saranya, N., Sudha, V., & Sweetha, S. (2017).

Assessment of Intracardiac Masses Classification. Journal of Chemical and Pharmaceutical Sciences, 5,101–103.

[22]. Arulmurugan, R., and H. Anandakumar. "Early detection of lung cancer using wavelet feature descriptor and feed forward back propagation neural networks classifier."

In Computational Vision and Bio Inspired Computing, pp. 103-110. Springer, Cham, 2018.

[23]. Thabsheera, AP Ayshath, T. M. Thasleema, and R. Rajesh. "Lung cancer detection using CT scan images: A review on various image processing techniques." In Data Analytics and Learning, pp. 413-419. Springer, Singapore, 2019.

[24]. Yu, Haixin, Zhong Guan, Katarina Cuk, Yan Zhang, and Hermann Brenner.

"Circulating MicroRNA Biomarkers for Lung Cancer Detection in East Asian Populations." Cancers 11, no. 3 (2019): 415.

[25]. Bhattacharjee, Ananya, and SwanirbharMajumder. "Automated Computer-Aided Lung Cancer Detection System." In Advances in Communication, Devices and Networking, pp. 425-433. Springer, Singapore, 2019.

[26]. Adir, Yochai, ShovalTirman, Shirley Abramovitch, Cynthia Botbol, Aviv Lutaty, TaliScheinmann, EyalDavidovits et al. "Novel non-invasive early detection of lung cancer using liquid immunobiopsy metabolic activity profiles." Cancer Immunology, Immunotherapy 67, no. 7 (2018): 1135-1146.

[27]. Shim, Jina, Myonggeun Yoon, and Youngjin Lee. "Quantitative study of fast non- local means-based denoising filter in chest X-ray imaging with lung nodule using three-dimensional printing." Optik 179 (2019): 1180-1188.

(15)

[28]. Khairuzzaman, Abdul KayomMd, and SaurabhChaudhury. "Masi entropy based multilevel thresholding for image segmentation." Multimedia Tools and Applications 78, no. 23 (2019): 33573-33591.

[29]. Mirjalili, Seyedali, Amir H. Gandomi, Seyedeh Zahra Mirjalili, ShahrzadSaremi, HossamFaris, and Seyed Mohammad Mirjalili. "Salp Swarm Algorithm: A bio- inspired optimizer for engineering design problems." Advances in Engineering Software 114 (2017): 163-191.

[30]. Prabhakar, Telagarapu, KilpattuRamaniharanAnandh, and Annapantula Sudhakar.

"Classification of Alzheimer’s Condition in T1-Weighted MR Images Using GLCM and GLRLM Texture Features." In Proceedings of International Conference on Wireless Communication, pp. 533-541. Springer, Singapore, 2020.

[31]. Mafarja, Majdi, Ibrahim Aljarah, HossamFaris, Abdelaziz I. Hammouri, Al- ZoubiAla’M, and SeyedaliMirjalili. "Binary grasshopper optimisation algorithm approaches for feature selection problems." Expert Systems with Applications 117 (2019): 267-286.