A NON-INVASIVE METHOD FOR BRAIN TUMOR DETECTION USING COMPUTER VISION AND DEEP LEARNING TECHNIQUES

Abstract

In the brain tumor diagnosis technique, the study of the brain MR image is supportive. For human life, cancer and tumors are deadly and damaging diseases. In the world of the Biocomputing field, this study is an additional struggle to tell about the position of image classification. A brain tumor is an uncontrolled growth of cancerous and non-cancerous cells in the brain. A brain tumor may be malignant or benign. The brain tumor symptoms depend upon the size of the tumor, location, and its type. The brain tumor is classified into two different types: a secondary brain tumor and a primary brain tumor. The organs of the brain cell where the tumor is located and grows up from these cells; this type is called a primary brain tumor. A cell of the tumor that belongs to another part of the body and may extend rapidly into the brain is called a secondary brain tumor. To evaluate and check the competence of the suggested model, the MATLAB tool is used. In this research, the dataset is collected from the Bahawalpur Victoria Hospital (BVH). It is concluded that this proposed method is better than other existing methods in terms of computation time after analyzing the results. Properties of feature extraction, i.e., mean, entropy, standard deviation, variance, connectivity, and many other features, are obtained. I have set a central tendency value in mean, standard deviation, and variance. If the value is less than the central tendency, that refers to a primary brain tumor; otherwise, it will be a secondary brain tumor. I have compared the results from other proposed methods. our proposed technique gives better results with an accuracy of 92.93. In the future, further classifier techniques may reach to find better results. Brain tumors, especially glioma, meningioma, and residual tumors, are common with a low survival rate. Clinically, they are identified on MRI scans and are classified after invasive methods. Spinal tap and biopsy are the methods used to determine the type of brain tumor. In this research work, a CNN-based architecture has been proposed to classify brain tumors in a non-invasive manner. Three classes, i.e., glioma, meningioma, and residual tumor, have been classified. The dataset has been collected from Bahawal Victoria Hospital, Bahawalpur, Pakistan. Experiments have been performed in different ways: 1) processing of as it is images present in the dataset, 2) processing the tumor segmented images, and 3) processing the large number of tumor segmented images. Experiment no. 01, experiment no. 02, experiment no. 03, experiment no. 04, experiment no. 05 and experiment no. 06   have achieved the accuracy as 66.13%, 80.93%, 88.07%, 91.33%, 92.00%, and 92.93% respectively. The proposed method has achieved an accuracy of 92.93%, which is high compared to the state-of-the-art methods. It has been experimentally proven that increasing the number of images has increased the achieved accuracy.  In future work of this research, all other brain tumor types will be classified. It is further aimed that will four WHO grades will also be classified using a non-invasive method to replace biopsy and spinal tape methods.