Brain Tumor Detection System using Convolutional Neural Network

Brain tumors, in medical terms, are the intentional or unintentional growth of mass cells which hamper the conventional functioning of the shape of a brain. For correct diagnosis and efficient treatment planning, it is necessary to detect the brain tumor in the early stages. The tumor within the brain is one of the most dangerous diseases and might be diagnosed easily and reliably with the assistance of detection of the tumor using automated techniques on MRI Images. Positron Emission Tomography, Cerebral Arteriogram, spinal tap, and Molecular testing are used for tumor detection. Digital image processing plays an important role in the analysis of medical images. Segmentation of tumors involves the separation of abnormal brain tissues from normal tissues of the brain. Over the few past years, various researchers have proposed semi and fully-automatic methods for the detection and segmentation of Brain tumors. The motivation behind the paper is to detect neoplasm and supply the better treatment for the suffering. The objectives of the paper are to develop an end-product (Web Application) that can be installed at hospitals. To facilitate this a detection model is developed that may accurately predict if an uploaded MRI scan of the brain shows it is affected by a tumor or not. To implement the paper a Convolutional Neural Network(CNN) was used to define the model. Transfer Learning is implemented to efficiently train the model. The data set used is split into 3 sets which are train, test and validation, in the ratio 80:10:10. The model is meant to be trained for 12 epochs. Callbacks also have been given to automate the model save process. The test accuracy of 97% is achieved. This trained model will be connected with an online Application via API. Within the proposed Web App the user is having access to four routes; which is a welcome page and which contains information about the system, the second route is information and awareness about the brain tumor in medical terms, third is the detection page, in which the trained model is deployed. The user can provide an input image, MRI images in our case, and the last route is the team information. Images which are fed to the model route will be processed by the developed convolutional neural network which can then confirm if a tumor is present or not and intimidate the user for the same through an output Display. The advantage of using this system is that it will automate the detection process, and ease the workload of the hospital staff. However for the advantage to become a reality, careful selection of accurate data is needed, or else there is a chance of false results.

I. Introduction

Computer vision and image processing are a few of the most significant domains in upcoming technical advancements of the world. The main objective of computer vision and Image processing is to enable machines and devices to view and capture the world in the way humans see it. It is the entire process involving object detection, classification and analysing of the results. Image processing has majorly two parts involved in the entire process. Pre-processing is the first step in image processing which consists of operations such as image enhancement, resizing, and adjusting images. Post Processing comes as the second part of the process which has special modifications as per the needs - highlighting the segmented areas, removing noise, and applying texts to the area needed. Thus, any image dataset which consists of similar fields can be processed and modified for particular needs to tackle the problems.