SSO-RBNN driven brain tumor classification with Saliency-K-means segmentation technique
Graphical abstract
Introduction
Brain tumor is a well known deadly diseases where early detection of tumor is a crucial aspect of getting cure. An unusual and uncontrolled growth of cells in human brain is termed as brain tumor which leads to drastic variations in brain behavioral pattern. It causes cancer which may yield around 13% of deaths worldwide. Moreover, statistically it has been found that 36% of people with cancerous brain tumor survive for 5 years and 31% survive for 10 years [1], [2], [3]. Tumors in brain are categorized into two different types such as malignant and benign. The tumor that contains cancerous cell is considered to be harmful and termed as malignant whereas the tumor with absence of cancerous cell is less harmful and termed as benign. The tumor size, style, location, behavior, and growth rate are the factors that show the level of risk in brain. These information helps to take decision for proper treatment like radiation, chemotherapy, and surgery. Accurate detection and localization of brain tumor in early stage provides a greater chance in survival of a person [4]. There exist various imaging techniques such as magnetic resonance imaging (MRI), computed, tomography (CT), single-photon emission computed tomography (SPECT), positron emission tomography (PET), Ultrasound, and X-ray to deal with brain tumors, Corona virus, and cancerous cells [5]. MRI is the most popular non-invasive technique which offers best contrast high spatial resolution-based image with cross-sectional views of tumor cells [6]. Detection and classification of tumor is an essential task to provide good treatment. However, medically it is risky and time-consuming stage in which a patient undergoes through few conventional process such as biopsy, manual inspection, expert suggestions etc. Therefore, it requires a Computer-Aided Design for early detection and classification of brain tumor without human intervene. For this, an accurate segmentation to locate tumor and efficiently classify it as normal or abnormal (tumorous) are the challenging issues.
Detection of tumor region from MR image depends upon proper segmentation whereas classification depends upon extraction of relevant features and an efficient classifier. There exist numerous techniques which are broadly categorized as brain tumor object segmentation and classification techniques. Tumor segment and feature extraction technique includes mostly different clustering-based segmentation algorithm, wavelet, PCA, intensity, texture, shape, LBP etc. [7], [8], [9], [10], [11]. Similarly, brain tumor classification technique used machine learning algorithm such as linear regression, K-nearest neighbor (KNN), Neural Network, support vector machine (SVM), random forest etc. are employed for detecting and classifying the tumor into normal or abnormal [12], [13], [14], [15], [16], [17]. Recently various convolutional neural network techniques are utilized such as CNN [18], [19], [20], [21], [22], [23], Generative Adversarial Network [24], [25], Deep Learning [26], [27], [28], and Transfer learning [29], [30] have been developed for segmentation with automatic feature extraction. However, these techniques take more time for execution and had no assurance of the accuracy which leads to an additional manual detection process to cross check the obtained results of classification models. Moreover, it leads to erroneous while dealing with bulk image data. To mitigate this issue, we formulate a brain tumor detection model including a new hybrid salience-K-mean based segmentation technique along with utilizing the advantage of social spider optimization (SSO) algorithm in Radial Basis Neural Network (RBNN) to yield significant performance.
The objective of the paper is to perform classification on the brain MR image by using proposed Saliency-K-mean-SSO-RBNN method which comprises preprocessing stage followed by segmentation and feature extraction, and finally classification. Initially, the MRI image is pre-processed to remove noise and convert it into a high contrast image. Then hybrid Saliency-K-mean is used for segmentation to determine the shape and position of tumor. Next, novel features are extracted by using Wavelet Transform (WT), PCA, kurtosis, skewness, IDM and DCT to yield a feature vector representation. Then the feature vector is applied into the RBNN to classify it by using an optimal cluster center which is further determined through SSO algorithm.
In this paper, the major contributions are listed as follows:
- •
We propose a new hybrid Saliency-K-mean based segmentation technique which detects the tumor region irrespective of its size with better accuracy.
- •
Novel features are extracted by using Wavelet Transform (WT), PCA coefficients for statistical characteristics, kurtosis for quantifying probability distribution of image texture, skewness for detecting the edges of a tumor by measuring the lack of symmetry, IDM for measuring local homogeneity, and DCT for zonal masking.
- •
We propose an efficient classification approach by combining SSO with RBNN where SSO discovers the optimal cluster center for RBNN with a high convergence speed and able to classify the brain tumors (Normal or Tumorous).
- •
Experimental results validate that the of proposed method Saliency-K-mean-SSO-RBNN achieves a better accuracy and computationally executes in a shorter time span (in seconds) as compared to existing methods.
The rest of this paper is structured as follows: Section 2 presents related works, Proposed approach is detailed in Section 3. Section 4 reports experimental result analysis and conclusion is in Section 5.
Section snippets
Related work
In present years, the detection of brain tumor from MR images are getting more challenging. An accurate detection of tumor region and prediction play vital roles in diagnosis. There exist a number of machine learning models for detecting the brain tumor which is further utilized for the development of automated system in medical field. The brain tumor detection process can broadly be categorized into brain tumor region segmentation and classification techniques as described in following
Proposed Saliency-K-mean-SSO-RBNN method
The proposed method comprises four basic steps such as pre-processing, segmentation, feature extraction, and classification. Fig. 1 depicts the overview of the proposed methodology.
SSO-based optimal cluster center
SSO algorithm follows the principles of swarm intelligent algorithm which is rooted with the cooperative aspects of social spiders [39], [40]. In multidimensional feature space, to find the sub-optimal center point, SSO is far better as compared to existing optimization algorithms such as Genetic Algorithm (GA), Particle swarm Optimization (PSO), Ant Colony Optimization (ACO). These existing algorithms may have premature convergence whereas SSO is capable enough to provide a balanced solution
Dataset
The proposed Saliency-K-mean-SSO-RBNN is validated on three standard datasets through experimentation. Table 1 details the description of the datasets. Dataset 1 [41] contains 296 images, out of which 152 images represent normal and rest 144 represent abnormal (tumor). Dataset 2 [42] comprises 250 images, out of which 100 images are normal and 150 images are abnormal. Similarly, Dataset 3 [43] has 369 images, where 155 image is abnormal and 214 is normal. A few classified normal and abnormal
Conclusion
In this paper, proposed Saliency-K-mean-SSO-RBNN is designed for classification of brain MR image into tumor and non-tumor. Initially, the modality of the given input image is pre-processed to enhance quality of images by removing the additional noises and convert it into gray image. Then in the next stage, these images are used for segmentation to separate tumor region using new hybrid Saliency-K-mean. Feature extraction is performed after segmentation to extract novel features such as Wavelet
Abbreviations
ANN : Artificial Neural Networks CT : Computed Tomography CNN : Convolutional Neural Network DCT : Discrete Cosine Transform DWT : Discrete Wavelet Transform GAN : Generative Adversarial Network IDM : Inverse Difference Moment KNN : K-nearest neighbor LBP : Local Binary Pattern MCC : Matthews Correlation Coefficient MRI : Magnetic Resonance Imaging PCA : Principal Component Analysis PET : Positron Emission Tomography RBNN : Radial Basis Neural Network SPECT : Single-Photon Emission Computed Tomography SSO : Social Spider Optimization SVM :
CRediT authorship contribution statement
Aparajita Nanda: Conceptualization, Validation, Investigation, Resources, Writing – original draft. Ram Chandra Barik: Software, Formal analysis, Investigation, Data curation, Writing – review & editing, Visualization. Sambit Bakshi: Methodology, Formal analysis, Investigation, Data curation, Writing – review & editing, Supervision, Project administration, Funding acquisition.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This research is supported by the following projects:
- a.
Project titled “Deep learning applications for computer vision task” funded by NITROAA, India with support of Lenovo P920 and Dell Inception 7820 workstation and NVIDIA Corporation with support of NVIDIA Titan V and Quadro RTX 8000 GPU.
- b.
Project titled “Establishment of Bioinformatics and Computational Biology Centre: Animal Bioinformatics - BIC at National Institute of Technology Rourkela” by Department of Biotechnology, Ministry of Science
References (45)
- et al.
Brain tumor detection in MR image using superpixels, principal component analysis and template based K-means clustering algorithm
Mach. Learn. Appl.
(2021) - et al.
A novel end-to-end brain tumor segmentation method using improved fully convolutional networks
Comput. Biol. Med.
(2019) - et al.
Multi-grade brain tumor classification using deep CNN with extensive data augmentation
J. Comput. Sci.
(2019) - et al.
GCAUNet: A group cross-channel attention residual UNet for slice based brain tumor segmentation
Biomed. Signal Process. Control
(2021) - et al.
Deep neural network with generative adversarial networks pre-training for brain tumor classification based on MR images
Biomed. Signal Process. Control
(2020) - et al.
Adversarial learning for deformable registration of brain MR image using a multi-scale fully convolutional network
Biomed. Signal Process. Control
(2019) - et al.
Optimization driven deep convolution neural network for brain tumor classification
Biocybern. Biomed. Eng.
(2020) - et al.
MR brain segmentation based on DE-ResUnet combining texture features and background knowledge
Biomed. Signal Process. Control
(2022) - et al.
Brain tumor detection using fusion of hand crafted and deep learning features
Cogn. Syst. Res.
(2020) - et al.
Brain tumor classification using modified local binary patterns (LBP) feature extraction methods
Med. Hypotheses
(2020)