Brain Tumour Detection Using CNN Sri lekha Jagannadham 1* , K. Lakshmi Nadh 2@ , M. Sireesha 3# Department of CSE, Narasaraopeta Engineering College, Narasaraopet, Andhra Pradesh 1 srilekhasunny@gmail.com, 2 lakshminadh37@gmail.com, 3 sireeshamoturi@gmail.com Abstract - The perilous disease in world nowadays is brain tumor. Tumor will occur when the healthy tissues are damaged and affects the brain. Tumor is the unlimited growth of bizarre cells in brain. Hence, death will be caused when there is a rapid growth of tumor cells. Detecting the abnormal tissues from normal brain tissues is the pivotal role for brain tumor detection system. M ain thing is that the concepts of medical image processing, M R images, these are mainly utilized by domain of brain tumor analysis. Detecting and diagonising the brain tumor in early stage is the significant task which can save a patient from worse effects. In our work, the input is MRIs (Magnetic Resonance Images) and from the input this research work attempts to extract tumor cells. Some of the pre-processing techniques of deep learning are used for removal of noise from the input images, to make the results more accurate, this research work will apply augmentation techniques to increase the training set and apply different Convolution Neural Network [CNN] techniques to grab out the best details from the image. In order to enhance the details from the image, the outer portion of the image cut out from all 4 sides. Keywords: Tumor, bizarre cells, Magnetic Resonance Image. 1 INTRODUCTION Brain tumor is one of the rigorous diseases in the medical science domain. Group of cells are abnormal and they are formed from uncontrolled division of cells, which is also called as tumor and these are formed and spread to spinal cord and different cells of brain [14][26]. Main key concern of a radiologist is the effective and efficient analysis of premature phase of tumor growth. The tumors are assorted into two, benign and cancerous. If tumor is not properly diagnosed and treated there is a chance of causing death [2]. The tumor is an unusual growth of tissues and uncontrolled cells and its rise [27]. Based on the stereotactic biopsy test histological grading is the gold standard and convention for detecting the grade of brain tumor. In biopsy procedure neurosurgeon will drill a small hole into skull where the tissue is collected. Biopsy test has many risk factors, the factors are bleeding from tumor and brain causing infection, seizures, severe migrane, stroke, coma and even death. Stereotactic biopsy’s main concern is that it is not 100% accurate and also result in serious diagnostic error followed by wrong clinical management of the disease. Using MRI, this research work obtains the brain images, and it can also recognise the noise and any changes using these MRI’s during acquisition [16]. MRI (Magnetic Resonance Imaging), it is an imaging technique and extensively used for diagnose the patient and treatment of tumors in clinical practice [18][20][22]. The image contains non-invasive soft tissue and this is used for diagnosis of tumors within the brain [1]. MR images are taken in three directions and they are sagittal, axial and coronal. MRI contain a noise caused by operator performance which can lead to serious inaccuracies classification [21]. The users can learn the pattern of brain tumor using deep learning techniques because manual segmentation is time consuming and also being susceptible to human errors or mistakes. Computer detection systems are challenging aspects in every field and still there is an open problem because of difference in shapes, areas, and sizes of tumor [3]. Techniques which are mentioned above will give segmented MRI without restricting the region of tumor; some of the techniques [24] are able to detect the single tumor but no technique will address the localization and detection of very small tumors. MRIs are majorly used to detect and visualize the details in internal structure of body [25]. In our paper, we proposed an automatic technique that detects the multiple and also very small tumors. The contribution and novelty of proposed system is that, it can detect localize and multiple tumors and small tumors which are not recognized in manual segmentation and the proposed model uses the same MRI image which is used to identify the tumor manually no need to change of that images. The segmentation should separate the active tumorous tissue from the necrotic tissue, and also the edema (swelling near the tumor) should be identified. This process is done by identifying the areas which are abnormal when compared to normal tissue [6]. When applying the classic segmentation methods and limitations such as inhomogeneous intensity, complex physiological structure and blurred tissues boundaries in brain MR images usually lead to unsatisfactory results [17]. So, we are using 2 different Activation functions to build the model accurately and different layers for creating CNN. The Activation functions are Relu and Sigmoid. To get the better results we are using 200 epochs to train the model. During each epoch training, it’ll save the model file in the directory. We can choose the best model to predict the results for the test data. 2 LITERATURE SURVEY Many researches has shown that, they worked on image processing and soft computing has a review and analysis on detection and augmentation[28] of brain tumour techniques. There are many other researchers working on the brain tumour detection techniques[29] because it is one of the most searching and challenging task. At present segmentation based on Neural networks became more prominent and increasing each and everyday The process of segmenting based on the Morphological operations and SFCM algorithm which also makes the computation time more. Our model(proposed) detects the tumour with almost 92% accuracy. This process of segmenting with MMO is established by Devkota [7]. Based on the segmentation [26] of histogram technique on the detection of edge approach is processed by Yanatao[8]. Dina [11] has introduced PNN [24] model, which is related to Vector Quantization. Some other authors has also implemented the PNN on segmenting technique along with PCA, which can reduces the dimensionality and helps in improving the feature extraction. Proceedings of the Fifth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) IEEE Xplore Part Number: CFP21OSV-ART; ISBN: 978-1-6654-2642-8 978-1-6654-2642-8/21/$31.00 ©2021 IEEE 734 2021 Fifth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) | 978-1-6654-2642-8/21/$31.00 ©2021 IEEE | DOI: 10.1109/I-SMAC52330.2021.9640875 Authorized licensed use limited to: Indian Institute of Space Science And Technology. Downloaded on March 25,2022 at 04:26:36 UTC from IEEE Xplore. Restrictions apply.