Comparative Study of Efficient Methodology for Tumor Detection
S. Gayathri1*, Dr. E. S. Madhan2, Dr. J. Avanija3
1* - Assistant Professor, Information Technology, Karpagam College of Engineering, Tamilnadu, India
2 – School of Computing, SRM Institute of Science and Technology, Chengalpattu, Tamilnadu, India
3 – Associate Professor, Department of CSE, Sree Vidyanikethan Engineering College, Tirupati, India
ABSTRACT
Tumor detection in human plays a vital role to be identified at an early stage and with more accuracy of the tumors. Tumors can be detected using many imaging techniques available.
However, computer aided algorithms can be used on images for effective classification of tumors and it‟s stages. Using any one of the efficient computer aided technique, it is easy to segment abnormal tissues in the brain with more accuracy. With the help of these techniques the tumor in the brain can be detected. The MRI images are taken, preprocessed and fed into the algorithm for detection of tumors after segmentation process. From the segmented results, features are extracted and classified accordingly based on the classification techniques. For this purpose many algorithms to detect the tumor cells have been proposed. In this paper a comparative study was done to find out the efficient methodology to detect the tumor cells using MRI with more accuracy.
Keyword: Hybrid CNN, Segmentation, Tumor Detection
INTRODUCTION
Tumor is an abnormal tissue growth where in the cells gets divided enormously [1]. It can be located on any part of the body. The malignant tumors are very dangerous because they can cause cancer if persists without any care. So these tumors should be identified and treated at an early stage. Tumors are classified into many categories like Grade I, II, III … If the tumor is identified at an initial stage which means the cell division has just then started and not of a larger area then this comes under grade I brain tumors. If these tissues are completely removed by a surgery then this stage tumor can be cured. In the next level the tumor starts growing slowly and its spread is much lesser to other regions. This type of grade II tumors can be treated but not completely removed by surgery. Grade III and others -The spread is higher and rapid in the nearby tissues and there may be a chance of recurrence. Usuallythese typesof tumors cannot be cured. Detecting brain tumor is highly complicated task and needs more effort [4] Diagnosis of brain tumors in the early stages of the tumor‟s start is difficult because it cannot accurately measure the size and resolution of the tumor [5][18]. So, the detection and treatment of tumors at
an early stage of tumor formation process may results in the patient‟s treatment and cure.
Therefore, the treatment of tumor depends on the timely diagnosis of the tumor [6]. MRI images provide better results than other imaging techniques such as Computed Tomography (CT), due to their higher contrast in soft tissue in humans [7] [15].By inspecting the MRI the tumors can be differentiated and they can be treated. In the year 2016, a study says that brain tumor related death is increasing in children among the age group 0–14 in the United States and they are deadlier than Leukemia [8]
In earlier days where there was no machine intelligence[19][23][24] theses detection of brain tumor happened by Expert opinion with multiple persons discussion and suggestion, biopsy(which is also risky), Human inspection, etc.[9] [16]. After a suggestion given by an expert, the same will be considered by other experts to make decision, because human brain is always not correct and the chances of misconception is higher because human eyes cannot diagnose most of the tumors and miscalculation may lead to serious health problems in patients..
It is clearer that the manual procedures are time consuming to classify tumor correctly which further delays treatment of tumor. With the advancements in the field of machine intelligence (computer vision) computer-aided diagnosis are widely used in many biomedical fields for accuracy. So the need of a high resolution technique that can detect the presence of tumors in brain is needed for experts to proper diagnosis among patients.
The major objective of detection of Tumors in brain image is carried out by carefullyextracting the features and processing it for segmentation and identifying the tumorous tissues [10][22][25]. Various methods have been proposed for detecting tumors. So a fully automated methodology [17] [20][21] for detection of brain tumor is required. The ultimateaim is to understand various computer aided methods or approaches that are developed for the detection of Brain tumors and their accuracy. This are compared may help to identify the efficient technique and that methodology can be used in the medical field for better performance of detecting tumors.
METHODOLOGY
Various methods have already been proposed and their flow of work in the process of detection of tumors is as follows. Generally every methodology undergoes these stages for detecting tumors.
A. Acquisition of MRI
This step is the initial step that has to be done for classification of tumors. Without a proper data set collection, classification cannot be done effectively. So, for this purpose various images of brain are collected. Few dataset are collected from various hospitals and few others have collected from the universally accepted databases.
B. Pre-Processing
The very first step that has to be done in all methodology is pre-processing of the images.
This stage helps to remove the unwanted noises, bias field distortion and motion heterogeneityetc. In this stage the images get enhanced and the results produced from these images will be better than these images that are not undergone this stage.
Even if the image was taken from a high resolution devices or any perfect database set , they have to be preprocessed in order to enhance contrast for better segmentation. Noises are removed using filters like low pass, Gaussian filter, median filter etc accordingly. A main concern in preprocessing of MRI is that it should be done carefully so as to maintain the standard or else poor result production will occur due to increased noise or important details being eliminated during the noise reduction.
C. Extraction of Features
The process of getting the useful information from a larger data is called Feature Extraction. By doing this the results can also be improved. Various features like colour, texture and geometric features which are more relevant for the detection of tumors are obtained from the feature extraction stage from themedical images. This help to reduce the number of dimensions needed for processing where the loss of relevant information does not occur.In this section the region of interest is extracted from the image using any of the segmentation technique. Energy,
Collection of brain images from various patients
Preprocessing techniques
Segmentation
Classification of images
Entropy, Correlation, contrast are the characteristics obtained from the image using the technique.
Energy = 𝐼(𝑖, 𝑗)𝑖,𝑗 2 (1)
Entropy = 𝐼(𝑖, 𝑗)log(𝑖 − 𝑗)𝑖,𝑗 (2)
Correlation = 𝐼 𝑖,𝑗 𝑖−𝜇𝑖 (𝑖−𝜇 )
𝑖,𝑗 𝜎𝑖𝜎𝑗 (3)
Region based segmentation, Segmentation using thresholding, watershed segmentation, Convolutional Neural Network (CNN) is used for learning how to segment images. CNN extracts features directly from pixel images with minimal preprocessing. CNN with softmax, RPF, Two layer CNN, CNN with region growing based segmentation are also used for feature extraction.
D. Classification
From the extracted features classifiers are used to classify the type of tumor and its level can be identified from the results. Various methods are proposed and their results have been discussed at the end of this paper.
E. Performance Evaluation
The parameters used to measure the performance of the classifiers are accuracy, sensitivity and specificity. Global performance can be identified using Accuracy. Reliability of the system is measured using Sensitivity and specificity measures the reliability of the system at making negative identifications.
Accuracy = 𝑇𝑃+𝑇𝑁
𝑇𝑃+𝐹𝑃+𝐹𝑁+𝑇𝑁 (4)
Sensitivity = 𝑇𝑃
𝑇𝑃+𝐹𝑁 (5)
Specificity = 𝑇𝑁
𝐹𝑃+𝑇𝑁 (6)
Where,
TP, TN, FP, FN represents True Positive, True Negative, False Positive and False Negative respectively
E. Process flow
Step 1: Input images are collected from various resources
Step 2: Pre processing technique are applied and enhanced using enhancement technique.
Step 3: Edge detection or background feature extraction are done Step 4: Classifiers are used to classify the type of tumor
Step 5: Finally the tumors are detected and send with tags mentioning their types
RESULTS AND DISCUSSION
The results produced by various methodologies are discussed here and tabulated for better convenience.
Segementation
Methodology Results
Precision Specificity Sensitivity
Hybrid 83% 89% 98%
CNN+ Softmax 97.34% 94.64% 100%
CNN+ RBF 94.24% 89.28% 100%
2 layer CNN 98.51% 84.19% 100%
CNN without any feature extraction methods
98.67% 96.42% 100%
Patchbasedapproach - 90% 94%
Classification
Methodology Accuracy
Segmentation using Thresholding 91.34%
CNN+ Softmax 98.67%
CNN+ RBF 97.34%
Region based Segmentation 87.48%
2 layer CNN 94.68%
CNN without any feature extraction methods 99.12%
Patchbasedapproach 98.99%
Watershed Segmentation 92.76%
CONCLUSION
Automated classification system became an important technique in computer-aided diagnosis for effective tumor detection among patients. Instead of going into a single approach a hybrid can be used for improved efficiency and efficacy for brain tumor segmentation. The accuracy of various methodsare discussed and the results can be seen in the above table.
Considering the value of diagnosis the medical accuracy of the method should be considerably high using the fully automatic computer aided methodologies.
REFERENCES:
[1] Images Zahra Sobhaninia et al, Brain Tumor Segmentation Using Deep Learning by Type Specific Sorting of Images
[2] Otsu, N., “A threshold selection method from gray-level histograms”, IEEE Transactions on Systems, Man, and Cybernetics, 9(1), 1979.
[3] Bhide, P. A. S., Brain Segmentation Using Fuzzy C means Clustering to Detect Tumour Region, 1(2), 85–90, 2012.
[4] Masoumeh Siar , Brain Tumor Detection Using Deep Neural Network and Machine Learning Algorithm, 9th International Conference on Computer and Knowledge Engineering (ICCKE 2019), October 24-25 2019, Ferdowsi University of Mashhad
[5] KS. Aboody, A. Brown, et al, Neural stem cells display extensive tropism for pathology in adult brain Evidence from intracranial gliomas, Proceedings of the National Academy of Sciences, 97 (23), pp. 1284612851, 2000.
[6] A. joshi, D. H. Shah, et al, Survey of brain tumor detection techniques through MRI images, International Journal of Research in Engineering and Technology, 10, pp. 586-594, 2013.
[7] JP. Poonam, Review of image processing techniques for automatic detection of tumor in human brain, International Journal of Computer Science and Mobile Computing, 2(11), pp. 117- 122, 2013.
[8] Gordillo, N., Montseney, E., Sobrevilla, P. (2013). State of the art survey on MRI brain tumor segmentation. Magnetic Resonance Imaging, 31(8), 1426–1438.
[9] Neha Baraiya and Hardik Modi ,Comparative Study of Different Methods for Brain Tumor Extraction from MRI Images using Image Processing, Indian Journal of Science and Technology, Vol 9(4), DOI: 10.17485/ijst/2016/v9i4/85624, January 2016
[10] Sidra Sajid1·Saddam Hussain2·Amna Sarwar1, Brain Tumor Detection and Segmentation in MRImages Using Deep Learning, Arabian Journal for Science and Engineering https://doi.org/10.1007/s13369-019-03967-8.
[11] Suganya, R., & Shanthi, R. , Fuzzy C- Means Algorithm - A Review, 2(11), 1–3. [11]
Kochar, P. (2014). A Survey on Brain Tumor Detection and Classification system based on Artificial Neural Network, 90(18), 16–21, 2012.
[12] Khawaja, A., Contrast Enhancement Impact on Detection of Tumor, 27(3), 2161–2163, 2015.
[13] Patil, A. J., et al., Pachpande, Automatic Brain Tumor Detection Using K-Means, 13896–
13903.
[14] Tanuja, M., & Shewale, P., Detection of Brain Tumor Based On Segmentation Using Region Growing Method, 5(2), 173–176, 2016.
[15] Madhumitha Ramamurthy , Y. Harold Robinson , S. Vimal , A. Suresh, Auto encoder based dimensionality reduction and classification using convolutional neural networks for hyperspectral images, Microprocessors and Microsystems , Volume No. 79, 103280, 2020.
[16] Madhumitha Ramamurthy, Ilango Krishnamurthi, S.Vimal, Y.Harold Robinson, “Deep Learning based genome analysis and NGS – RNA LL identification with a novel hybrid model”, Biosystems, Volume No. 197, 104211, 2020.
[17] Madhumitha Ramamurthy, Ilango Krishnamurthi, Sudhakar Ilango, Shanthi Palaniappan,
„Discrete Model based answer script evaluation using decision tree rule classifier‟, Cluster Computing, 22, 13499–13510 Nov(2019)
[18] G.Selva Marry, S. Manoj kumar “Self-verifiable Computational Visual Crptographic Protocol for Secure 2D Image Communication” International Journal of Measurement Science and Technology, Vol. 30, No. 12, 2019.
[19] G. Selva Marry, S. Manoj kumar “Secure grayscale image communication using significant visual cryptography scheme in real time applications” SPRINGER International Journal of Multimedia Tools and Applications, Pages 1-20, 2019
[20] S. Manoj Kumar N.Rajkumar “SCT Based Adaptive Data Aggregation for Wireless Sensor Networks” International Journal SPRINGER - Journal of Wireless Personal Communication Volume 75, Issue 4 April 2014, Page 2121-2133
[21]Ponmagal, R.S., Karthick, S., Dhiyanesh, B. et al. Optimized virtual network function provisioning technique for mobile edge cloud computing. J Ambient Intell Human Comput (2020).
[22]Ramamoorthy, S., Ravikumar, G., Saravana Balaji, B. et al. MCAMO: multi constraint aware multi-objective resource scheduling optimization technique for cloud infrastructure services. J Ambient Intell Human Comput (2020).
[23] Basha, A.J., Balaji, B.S., Poornima, S. et al. Support vector machine and simple recurrent network based automatic sleep stage classification of fuzzy kernel. J Ambient Intell Human Comput (2020)
[24]Balaji, B.S., Balakrishnan, S., Venkatachalam, K. et al. Automated query classification- based web service similarity technique using machine learning. J Ambient Intell Human Comput (2020)
[25]Viji, C., Rajkumar, N., Suganthi, S.T. et al. An improved approach for automatic spine canal segmentation using probabilistic boosting tree (PBT) with fuzzy support vector machine. J Ambient Intell Human Comput (2020).
