Article Preview
Top1. Introduction
Uncontrolled growth of cells in the brain causes brain tumors. Brain tumors may be cancerous or non-cancerous. A cancerous brain tumor is known as a malignant tumor and a non-cancerous brain tumor is known as a benign tumor (Sridhar et al., 2013). A roughly calculated statistic shows that only in the United States around 7,000,000 people have been affected primarily in brain tumors. Among them 69.8% are cancerous and 30.2% are non-cancerous. Moreover, around 87,000 people will be predicted as in 2020, of them, 61,430 will have cancerous and 25,800 will have non-cancerous tumors. The statistics also show that for the patients containing the malignant tumor, the mean progress rate is 36%. The statistics also predicted that due to malignant tumors around 18,020 people will die in the year 2020 (American Cancer Society, 2020).
Medical imaging techniques provide the first preface for the identification of brain tumors. Magnetic resonance imaging (MRI) is the first choice for this purpose. A magnetic field is used by magnetic resonance imaging (MRI) to generate overpowering images than CT scans (Sharma et al., 2014). MRI identifies the presence of a tumor in a more preferred way than other techniques. In T1-weighted MRI the brain tumor arrives with the same brightness as brain tissue but in T2-weighted MRI the tumor arrives with high brightness than brain tissue (Abbasi and Tajeripour, 2017; Tandel et al., 2019). Once MRI shows that there is a tumor in the brain, then we need to find the type (malignant or benign), size, and region of the tumor for its treatment.
Several techniques have been proposed for the diagnosis of brain tumors. Paper (Zhang et al., 2015) proposed a brain tumor identification and classification technique using a deep CNN with a three-plane structure. The system obtained 97% accuracy with a learning rate of 0.001. Paper (Banerjee, 2018) implemented a method to do classification and screening of brain tumors. Total 3486 images of brain MRI from two datasets were used in this method. A CNN was used to process the MRI slices and a recurrent neural network was used for classification. This method provided 92.13% accuracy. In a paper (Zhou et al., 2018) presented a method that used a deep learning technique to detect and classify brain tumors. The system provided 97.18% accuracy. In (Ari and Hanbay, 2018), fuzzy C-means clustering was used for segmentation of the MRI image, and discrete wavelet transform (DWT) was used for feature extraction. In addition, PCA (principal component analysis) based feature reduction technique was used to select important features. After that deep neural network (DNN) was used for classification. This system obtained an accuracy of 96.97%. In the paper (Mohsen et al., 2018), DWT and Gabor filter was applied on each manual segmented MRI slice to do feature extraction. The system used a back-propagation neural network to do feature classification. This process can identify three types of tumor (Meningioma, Glioma, and Pituitary) with an accuracy of 91.9%. 98% accuracy for brain tumor classification was obtained by a CNN using the segmented image through the watershed algorithm in paper (Wang et al., 1999).
The key issues in brain tumor detection are as: (i) extraction of potential details in MR images, and (ii) digging the most significant features. Our proposed method is to enhance and extract the informative region to grasp the edges and curves from the MRI. This can be suitably addressed by using the Curvelet and CNN. In addition, optimization of the potential features by redundancy minimization of the concatenated textural and deep features might reveal better outcomes by a classifier.