Deep Learning Techniques on Brain Images

Abstract

Our brain is the most composite organ of the human body with an aggregation of 100 million nerves which communicate in a nexus of synapses. All the activities from thinking, memorizing, storing information, the functioning of other organs of the body are all managed by the brain. Any disease which affects the brain affects the whole of the body. Severe brain diseases paralyze the body. Some of the common categories of brain diseases are seizures, trauma, tumor, and infections. Alzheimer’s, Epilepsy, Brain Cancer, and brain disorders. Research to use Image Processing Techniques in the field of brain diseases still has a long way to go. This paper is one such small step in the process of understanding Deep Learning in brain imaging. It is a detailed study on brain diseases and how algorithms can help in the current treatment. CNN is discussed in detail with its architecture and the reason of its popularity is discussed. This particular paper also comprises a case study of one such disease i.e brain tumor and the effect of various parameters in improving the accuracy of Convolutional neural networks on this particular data-set. The case study involves augmenting the data and applying CNN on it. The effect of CNN is then studied on the basis of three parameters which are Optimizers, Activation Function, and Loss Function. A comparative analysis is then drawn out among all the possible combinations and the best combination of these parameters are found. The models were evaluated in terms of accuracy and time required to train the algorithm. Using the comparison table important findings and conclusions were drawn out.