11 - Detection of Brain Tumor with Magnetic Resonance Imaging using Deep Learning Techniques

Abstract

Brain tumors are instigated by abnormal growth of tissues in brain due to uncontrolled and rapid cell division. This can be fatal if not promptly treated in earlier stages. Magnetic Resonance Imaging (MRI) scans are widely used in clinical practices for detection and treatment process. In recent times, Deep Learning (DL) techniques play a vital role in computer-aided diagnosis which has revolutionized the pace and accuracy of the diagnosis. As detecting brain tumor requires utmost accuracy and sensitivity, DL paves way for improving the accuracy, sensitivity, and specificity through its various techniques. One such technique is segmentation, which is mainly used to focus the particular area of infection. MRI segmentation when manually done is tedious, time-consuming process, and prone to human errors, which can turn disastrous while identifying brain tumors. DL's U-Net model does this quickly for MRI scan images and produces the segmented output. These segmented output images can be evaluated using dice coefficient to avoid misleading errors and to improve the performance. Finally, once segmented images are obtained, we can train and test them using various DL classification models such as Convolution Neural Network (CNN), Resnet50, AlexNet, VGG16, etc., and their performance metrics are evaluated. The best performing model would then be suggested to clinical practitioners for deployment in patient care.