BRAIN TUMOR IDENTIFICATION ENABLED USING ROBUST RF CLASSIFIER FOR IMPROVED MRI BRAIN IMAGES

BRAIN TUMOR IDENTIFICATION ENABLED USING ROBUST RF CLASSIFIER FOR IMPROVED MRI BRAIN IMAGES

Abstract

The diagnosis of Brain tumor identification is an essential and challenging process by extracting the MRI brain images. The various assortments like as brain image shapes, the intensity of images, regions of an image, illuminations, contrast are the conceivable factors that define the brain tumor identification. MRI images play a vital role in finding a brain tumor. That needs to attain a better efficiency of performance, which improves the performance of the Computer-Aided Diagnosis (CAD) method. The MRI brain images addressed the limitations of brain tumor radiance, the contrast of the brain images. In this paper, propose a robust Random Forest (RF) classifier to enhance MRI brain images’ above inabilities. Here, the MRI input process for the preprocessing stage to remove the noise present in the image, edge detection using High Pass Filter (HPF). The preprocessed image move to the image segmentation and then move to the feature extraction for extracting and selecting the features of the MRI images using Independent Component Analysis (ICA) for non-radiance of MRI images. The obtained image moves to the Random Forest (RF) classifier to enhance the accurate prediction of brain tumor detection (normal/abnormal MRI images). Finally, the brain tumor area is to be detected accurately. The performance of the brain tumor identification experiments carried out using MATLAB and evaluates the better performance. The outcome of the results compared with the other conventional brain tumor identification techniques.

Keywords: Brain tumor detection, MRI images, HPF, ICA, RF.

1. INTRODUCTION

In recent years, brain tumor detection can evolve human life, whether benign or malignant. A brain tumor is a bunch of abnormal/affected cells in the brain, forming an uncontrolled condition to build a grouping of brain cells in the brain and nearby the cells. The brain tumor is an irregular shape of cells that might be filled by the solid/ liquid state. When a specialist evolves, the brain tumors that are never changing the brain of the lesion but due to the tumor size, which may changes. The brain tumors may have in large/small lumps. However, any type of brain tumor cells can form into a tumor. There are multiple types of tumors, such as origin, shape, and cell types. A brain tumor is an irregularity in the brain caused by brain cells developing and partitioning in an uncontrolled way. The brain tumor leads to generate a life-threatening problem; there is an early stage of prediction that enables the vital key component. In medical image processing, brain tumors’ identification plays a crucial role in processing the exact outcome. Brain tumor detection has some challenging factors while diagnosing the MRI brain images. The brain image clarifies the benign and malignant of the MRI images.

The various stages involve specifying the particular area of the affected brain images in (Özyurt, Sert, Avci, & Dogantekin, 2019). The brain tumor detection and classification in MRI images play the most significant role and tough in image processing. In contrast, MRI images provide the compact form of images to diagnose the accurate results. The various preprocessing, feature extraction, segmentation, and classification approaches evolve the brain tumor detection discussed in (Rajesh, Malar, & Geetha, 2019). Brain tumor identification depends on the following characteristics a tumor growth rate and malignancy of the tumor when the brain tumor splits into two subdivisions, such as primary and secondary, like an early stage and peak stage. Based on the behavior, region, and origin of cells by WHO. The MRI images enable the better tumor diagnosis, which leads to no pain, non-invasive method, as well as generates the finely detailed information about the brain and its nerve cells discussed in (Sharif et al., 2020).

One of the most significant reasons for increasing morality like the reduced lifespan of the human leads to various diseases. Recent studies show that Brain tumor detection has the challenges as brain tumors in the brain are size variant, shape variant, shape variant, location variant, and image intensity variant. (Panda, Mishra, & Phaniharam, 2019). Machine learning algorithms like Neural Networks, Support Vector Machine and Convolutional Neural Network enable brain tumor classification. Here, the Random Forest (RF) classifier evolves the non-radiance, low contrast of brain tumor images. The optimized features allow the ICA to approach (Rajagopal, 2019). The researcher used to detect the brain tumor and track its progress in the treatment process. By using technology and the availability of MRI images, the researchers and the experts get a tremendous amount of information to make the decision related to a brain tumor. From high-resolution MRI images, it is easy to find the information related to brain structure and the organization of cells in the brain and from where abnormalities in the formation of tissues can be easily detectable  (Mittal et al., 2019). By consider the effective classification performance of brain tumor identification, the proposed Robust Random Forest (RF) classifier performs better than other conventional classification algorithms. RF classifier exploits the less radiation, higher contrast levels, and spatial resolution with the accurate classification prediction of MRI brain images. From this paper, the improved and enhanced performance of brain tumor detection using MRI brain images with the efficient classification of RF classifier approach.

The rest of the paper has been organized as follows: Section 2 gives a detailed analysis of the related works. Section 3 discusses the paper’s problem statement; Section 4 shows the proposed methodology with the detailed explanation, and Section 5 gives the results and discussion of the paper.

1. Related works

            (Amarapur, 2019) stated the brain tumor segmentation through the modified level set method in the glioma analysis. The MRI images can be preprocessed to remove the noise, and segmented through the level set algorithm innovatively. By segmenting the initial contour and intensity of the image pixel from the histogram equalization with the accuracy of prediction. (Zhang, Thapa, Haas, & Bastola, 2019) proposed the initialization of potential biomarkers to enable the dependable and reproducible biomarkers across the several expression of genes. To propose the gene expression through the Data-Driven Reference (DDR) approach to eliminate the platform based biases and the non-biological variable factors. (Wadhwa, Bhardwaj, & Verma, 2019) proposed the segmentation of brain tumors through the association of CRF with DeepMedic or Ensemble and Conditional Random Field (CRF) with Fully Convolutional Neural Network (FCNN) for effective brain tumor segmentation using the MRI brain images with start-of-art of quantitative analysis. (Perez & Capper, 2020) stated the Machine Learning (ML) through DNA methylation of brain tumor classification to retain the accurate results. The high‐quality molecular data through the DNA methylation from the specimens of formalin‐fixed paraffin‐embedded pathology. The additions of molecular testing enable the copy number analysis and identify the pediatric brain tumors by considering the DNA methylation process. (Kim et al., 2020) proposed the brain tumor images get extracted through the Radiomic features. The Radiomic model consists of ML-based feature extraction and Linear Model classifier. The concordance correlation coefficients (CCCs) help for the segmentation process, which is the temporal-based independent validation of MRI images.

(Abd-Ellah, Awad, Khalaf, & Hamed, 2019) stated the conventional machine learning techniques and deep learning techniques to validate the brain tumors. The brain tumor identification, segmentation, and classification are efficiently made with the three unique algorithms to attain better performance. (Kaplan, Kaya, Kuncan, & Ertunç, 2020) proposed the two unique feature extraction methods like nLBP and αLBP, to achieve the classification of brain MRI images. The formation of nLBP enables the brain tumor image pixels around the neighbors to allow the consecutive distance. The operator of αLBP depends on each pixel with its angle value with the high performance of classification and feature extraction using various classification algorithms. (Khan et al., 2020) stated the Internet of medical things (IoMT) computation method used to detects the brain tumor based images CT scans/MRI images. Here, the approach of Partial Tree (PART) for the usage of a rule-based association learner with the high accuracy with the less time consideration to attain the brain tumor detection. (Gudigar et al., 2020) stated an efficient brain pathology identification (BPI) algorithm that performs the human error reduction during the diagnosis of tumor detection. The rapid growth of the brain tumor identified through CAD tool.

(Mohan & Subashini, 2019) proposed an intelligent based medical imaging approach to evolved the leading-edge software packages, metrics validation, glioma evaluation. The medical images and the tumor diagnosis enable the exact accuracy of prediction. (Kalaiselvi, Kumarashankar, & Sriramakrishnan, 2019) stated the patch-based updated run-length region growing technique (PR2G) for useful classification and segmentation process. For classification, the SVM classifier to be utilized. An infinite feature selection (IFS) method used for feature selection and the run length region growing technique enabled the feature segmentation process to attain the normal/abnormal and find the tumor’s location. The experiments are done through the BraTS repositories and whole-brain atlas (WBA) to construct 3D tumor volume and the actual/segmented tumor volume. (Li, Guindani, Ng, & Hobbs, 2019) stated the clinicopathological based textural features derive from gray-level co-occurrence matrices (GLCM) to manipulate the context o cancer detection using CT scans. By considering the multivariate Gaussian spatial processes, the adrenal lesions to be evaluated. (Khamparia, Gupta, de Albuquerque, Sangaiah, & Jhaveri, 2020) stated a new internet of health things (IoHT) driven deep learning framework for brain tumor detection and tumor cell classification. The traditional Machine Learning (ML) techniques, namely KNN, RF, NB, LR, and SVM classifiers through the standard Pap smear Herlev dataset. (Valizadeh, Riener, Elmer, & Jäncke, 2019) stated the EEG based tumor detection by which it attains the higher bit rate compared to the BI quality parameters. By considering the comparison of the different algorithms, SVM and ANN retain the better prediction of brain tumor identification. The BI quality values are also higher for both of the classification methods.

III. Problem statement

The manual segmentation of abnormal brain tissues from the healthy brain cell consumes enormous time and can produce inaccurate results. To provide the solution for this and to help the clinical experts for the segmentation of brain tumor regions from MRI images, to create a new Computer-Aided approach to automate this process with the help of deep learning algorithms. In this paper, we have proposed a new plan that solves the medical image processing issue to detect the brain tumor with more accuracy.

1. Proposed Methodology

            The proposed methods overcome and satisfy the inabilities of the other conventional algorithms. The proposed approach’s performance enables the radiance of brain images to attain the contrast levels with better methodologies such as preprocessing, segmentation, feature extraction, and classification of MRI brain images. The schematic diagram of the proposed method shown in figure 1.

Figure 1 Schematic diagram of the proposed method

4.1 Preprocessing:

The brain tumor based images took from Magnetic Resonance Imaging (MRI). Before the image segmentation, the preprocessing stage of evolving. The preprocessing stage includes the filtering process. Namely, as Median Filter (MF) evolves the noise reduction of the gray level images in the brain images, fine details of the image, missing data replacement, and intensity variation by reducing the amount, thereby the variation in the neighboring pixels. The noise removal is an essential process in MRI pictures. Here, for noise removal median filter is examined. A median filter is a non-straight approach used for removal of noise present in the MR images. The median is evaluated by ordering all the pixel approval, and after that, displace the pixel consideration with the center of the pixel value. The mean filter incorporates the pixel to pixel variation by replacing each pixel value with the average of neighboring pixels includes itself also. The filter helps to evaluate the noise in the original MR images without sharpness diminishment conditions. Consider the input image I (a, b) size of M N, which consists of noise. Next, to separate the image into n number of levels L and each level has one center of the pixel value. Now, remove the noise and replace the center pixel through the use of median value evaluations. The equation of the median filter is given

I (a, b) = median _{(s, t)ϵ} {H (u, v)}                                                                            (1)

Where, Cxy is the set of rectangular sub-image levels, centered at the point (x, y). This process is repeated for the entire level present in the input image when the speckle noise to be removed. Finally, to obtain the noise-free image in the preprocessing stage. The preprocessed image can evolve the image segmentation to attain the normal/abnormal MR images according to its features.

4.2 Feature extraction:

Independent Component Analysis (ICA) helps extract the MRI brain images for the feature extraction stage. ICA incorporated the accumulation of brain image features contribution. ICA is an unsupervised learning technique because its performance results as the amount of maximum independent component features (f_a, f_b), which involves the uniform distribution such as [-1, 1] for binary feature extraction and the output performance (P) results denoted as given below:

P = O =                                                                                                            (2)

P is the binary feature extraction value to the nearby pixels (f_a, f_B) referred to as the grayscale values on the centralized pixel value. The obtained results show the decimal formation into the feature vectors. This ICA technique is connected to the dispersal contribution. The feature extraction pattern can enable independent components features of the obtained data pattern. ICA mainly used to extract “pure” details of images from the impure sources of brain MRI features. ICA degrades the obtained signal mixtures as an M × N matrix. The number of the signal source is well-known and less than the number of signal mixtures matrix, then the number of signals based MRI images get extracted by using Independent Component Analysis, which can be reduced by preprocessing signal source mixtures. For accurate detection of brain tumors through MRI, the MR image should define clearly between the multiple brain cells, such as white detail tracts and gray details. The traditional MRI images never display multiple tumor cells in detail. But, it is possible to use different MRI settings such that every setting captures a multiple image mixture of the source signals additional with the multiple brain tumor cells. Where M describes the measured mixture signals, N describes the number of variables. In matrix formation,

X = A.S                                                                                                                                   (3)

Where,

X is a non-linear vector with M*N elements

S is independent component vectors

A gets concealed mixing matrix

The A^-1 can calculate, and it referred to as W. The independent components vectors evaluated by finding an individual matrix W. The individual matrix W, which also called as demixing of the component matrix, can be calculated through an algorithm that optimizes the statistical independence of the component vectors of the mixture. If consider the two features for feature extraction

4.3 Classification:

An ideal hyperplane separation separates the MR images of two classes that also maximizes the hyperplane margin. For efficient prediction and classification of brain tumor MR images, the RF classifier performs better classification results. Random Forest classifier helps to classify the brain tumor features. Random forest classifier is an ensemble classifier that suits several decision trees on multiple subsamples of the MRI brain tumor images. Random forest performs better than other conventional classifiers, and also, there is no necessity for the number of classes to respond to the precise non-linear segment features. The ensemble method based on Each tree prepared to enable the non-linearly inspected information (features) with the equivalent alternatives from training samples of brain tumor images amid a specific time of learning. Various trees 380 are prepared to build the correlation and diminish the fluctuation between trees. In the test stage, each tree’s vote is considered, and the dominant part vote is given to the concealed information. RF is a potential analysis that provides an evaluation of errors and significant variables etc. Random Forest carried into a functional brain image classification within an association of time slots.

In the RF-enabled classification process, the features of MRI brain images used for training each association tree calculated the original MR images, not all the samples of training MR images. The non-linear selection of brain tumor images is essential for the classifier, namely RF classifier. When separating the nodes of each decision tree, the feature subspace is extracted instead of the better features of MR images in all the train sets. It separates the input MRI brain tumor image into subsequent nodes. The non-linear feature selection relieves the overfitting problems of RF classifier. By put to the test data, the MRI image-based pixel vector gets extracted in the same way and input to the trained model. As each decision tree in RF is independent of each other, the final prediction probability will generally be calculated by simply averaging the respective classification probabilities of all trees ‘t’ defined as:

P(y/F) =  (y/f (F, A)), t ϵ (1, T)                                                                                  (4)

Where P is referred to as the performance of classification.

F describes image intensity.

Y describes the label set.

An essential tunable parameter of Random Forest is the association rule-based tree depth, which can carry out to minimizes the errors that occurred in the technique.

1. Results and Discussion

The performance analysis of the brain tumor identification evolves through the MATLAB simulation. The parameters such as Structured Similarity Index (SSI), Mean Square Error (MSE), Standard Deviation (SD), PSNR ratio, and dice coefficient conditions. To attain the effective efficiency through sensitivity, specificity, and accuracy derived from the performance. Table 1 shows the performance analysis of the SSI, MSE, PSNR values.

1. Structured Similarity Index (SSI): SSI is a noncognitive metric that signifies the quality of image degradation that may be caused by compression/losses of MRI features in image processing. I

SSI = ()*()*()                                                                  (5)

SSI provides radiance levels, contrast, and structural content of the MRI based images.

1. Mean Square Error (MSE): It is the MR image fidelity calculation. The use of MR image fidelity measurement is to find the connection amongst two MR images through the better quantitative result. MSE calculated as,

SD () =                                                             (6)

1. PSNR: It is a measure to find the quality of the processed MR image.

PSNR = 20log₁₀                                                                                     (7)

Table 1 Performance analysis of SSI, MSE, and PSNR values

Here the True Positive (TP) is images that are appropriately acknowledged, and False Negative (FN) is the image incorrectly eliminated. True Negatives (TN) are the circumstances adequately reduced, and False Positives (FP) are the MR Images that are  recognized wrongly as brain tumor based MR Images.

1. Sensitivity: The proportion of ordered as true positives amongst the complete exudates of images.

Sensitivity =                                                                                                       (8)

2. Specificity: The capability to detect the portion of the population that does not have exudates that is true negatives and is stated as the proportion of ordered as TN amongst the complete non-exudates.

Specificity =                                                                                                      (9)

3. Accuracy: The degree of approximation of the measurement conforms to the corrected value.

Accuracy =                                                                                            (10)

Table 2 Comparison of accuracy with various parameters

The comparison of accuracy with various parameters shown in Table 2. The comparative analysis of Sensitivity, Specificity, and Accuracy analysis shown in figure 2. The Comparative analysis of TN, TP, FN, FP, shown in figure 3.

Figure 2 Comparative analysis of Sensitivity, Specificity, and Accuracy analysis

Figure 3 Comparative analysis of TN, TP, FN, FP

1. Conclusion

The brain tumor leads to the high-risk factor for humans, which is a life-threatening factor. The brain tumor detection in the image processing technique employs perfectly with the accurate detection of results whether the tumor is normal or abnormal through the MR Images. In this paper, the robust RF classifier used to employed the accurate prediction of a brain tumor in an early stage identification. Here, the various stages like preprocessing through HPF to remove the speckle noise in the grayscale level, feature extraction through the ICA by extracting the images according to its features, classification is done through RF classifier by classifying the normal/abnormal MR Image of the brain tumor cells. The binary patterns of the MRI to be filtered and retain the perfect edges get extracted in the extraction stage. The experiment carried out by using the simulation of MATLAB. The accuracy of the RF classification is 98.8%, which is more efficient than the existing IDSS based SVM classifier (97.02%). The sensitivity, specificity, TN, TP, FN, FP values get improved by comparison to the existing approach. By analyzing the result, the robust RF classifier based brain tumor detection actively enables the efficient brain tumor identification process.

References:

• Abd-Ellah, M. K., Awad, A. I., Khalaf, A. A., & Hamed, H. F. (2019). A review on brain tumor diagnosis from MRI images: Practical implications, key achievements, and lessons learned. Magnetic resonance imaging, 61, 300-318.
• Amarapur, B. (2019). Cognition-based MRI brain tumor segmentation technique using modified level set method. Cognition, Technology & Work, 21(3), 357-369.
• Gudigar, A., Raghavendra, U., Hegde, A., Kalyani, M., Ciaccio, E. J., & Acharya, U. R. (2020). Brain pathology identification using computer aided diagnostic tool: A systematic review. Computer Methods and Programs in Biomedicine, 187, 105205.
• Kalaiselvi, T., Kumarashankar, P., & Sriramakrishnan, P. (2019). Three-phase automatic brain tumor diagnosis system using patches based updated run length region growing technique. Journal of Digital Imaging, 1-15.
• Kaplan, K., Kaya, Y., Kuncan, M., & Ertunç, H. M. (2020). Brain tumor classification using modified local binary patterns (LBP) feature extraction methods. Medical Hypotheses, 109696.
• Khamparia, A., Gupta, D., de Albuquerque, V. H. C., Sangaiah, A. K., & Jhaveri, R. H. (2020). Internet of health things-driven deep learning system for detection and classification of cervical cells using transfer learning. The Journal of Supercomputing, 1-19.
• Khan, S. R., Sikandar, M., Almogren, A., Din, I. U., Fortino, G., & Guerrieri, A. (2020). IoMT-based computational approach for detecting brain tumor. Future Generation Computer Systems.
• Kim, M., Jung, S. Y., Park, J. E., Jo, Y., Park, S. Y., Nam, S. J., . . . Kim, H. S. (2020). Diffusion-and perfusion-weighted MRI radiomics model may predict isocitrate dehydrogenase (IDH) mutation and tumor aggressiveness in diffuse lower grade glioma. European Radiology, 30(4), 2142-2151.
• Li, X., Guindani, M., Ng, C. S., & Hobbs, B. P. (2019). Spatial Bayesian modeling of GLCM with application to malignant lesion characterization. Journal of applied statistics, 46(2), 230-246.
• Mittal, M., Goyal, L. M., Kaur, S., Kaur, I., Verma, A., & Hemanth, D. J. (2019). Deep learning based enhanced tumor segmentation approach for MR brain images. Applied Soft Computing, 78, 346-354.
• Mohan, G., & Subashini, M. M. (2019). Medical Imaging With Intelligent Systems: A Review Deep learning and parallel computing environment for bioengineering systems (pp. 53-73): Elsevier.
• Özyurt, F., Sert, E., Avci, E., & Dogantekin, E. (2019). Brain tumor detection based on Convolutional Neural Network with neutrosophic expert maximum fuzzy sure entropy. Measurement, 147, 106830.
• Panda, A., Mishra, T. K., & Phaniharam, V. G. (2019). Automated brain tumor detection using discriminative clustering based MRI segmentation Smart Innovations in Communication and Computational Sciences (pp. 117-126): Springer.
• Perez, E., & Capper, D. (2020). Invited Review: DNA methylation‐based classification of paediatric brain tumours. Neuropathology and Applied Neurobiology, 46(1), 28-47.
• Rajagopal, R. (2019). Glioma brain tumor detection and segmentation using weighting random forest classifier with optimized ant colony features. International Journal of Imaging Systems and Technology, 29(3), 353-359.
• Rajesh, T., Malar, R. S. M., & Geetha, M. (2019). Brain tumor detection using optimisation classification based on rough set theory. Cluster Computing, 22(6), 13853-13859.
• Sharif, M., Amin, J., Nisar, M. W., Anjum, M. A., Muhammad, N., & Shad, S. A. (2020). A unified patch based method for brain tumor detection using features fusion. Cognitive Systems Research, 59, 273-286.
• Valizadeh, S. A., Riener, R., Elmer, S., & Jäncke, L. (2019). Decrypting the electrophysiological individuality of the human brain: Identification of individuals based on resting-state EEG activity. NeuroImage, 197, 470-481.
• Wadhwa, A., Bhardwaj, A., & Verma, V. S. (2019). A review on brain tumor segmentation of MRI images. Magnetic resonance imaging, 61, 247-259.
• Zhang, L., Thapa, I., Haas, C., & Bastola, D. (2019). Multiplatform biomarker identification using a data-driven approach enables single-sample classification. BMC bioinformatics, 20(1), 601.
• Sharma, H., Govindan, K., Poonia, R. C., Kumar, S., & Wael, M. (2020). Advances in Computing and Intelligent Systems. Springer Singapore.
• Chen, G., Li, Q., Shi, F., Rekik, I., & Pan, Z. (2020). RFDCR: Automated brain lesion segmentation using cascaded random forests with dense conditional random fields. NeuroImage, 211, 116620.
• Polepaka, S., Rao, C. S., & Mohan, M. C. (2019). IDSS-based Two stage classification of brain tumor using SVM. Health and Technology, 1-10.