Brain Tumor Detection in Magnetic Resonance Imaging using Deep Learning Approach

In this paper, the design and implementation of a tumor detection system using two CNN models is considered. Digital image processing and Deep Learning technologies enable us to develop an automatic system for the diagnosis/detection of various kind of diseases and abnormalities. The tumor detection system may include image enhancement, segmentation, data augmentation, features extraction and classification; all these steps are discussed in details in the above sections. To work on CNNs, powerful GPU based system are required to speed up the process, lot of processing is carried out and also lot of RAM is required to process the images for testing. CNNs have also some options such as optimization technique selection, Number of Epoch, Batch size, iteration and learning rate. These options are tuned to get the optimal results from the CNN model. Learning rate is used to update the weights and bias in training phase, learning rate changes the weights. One Epoch is when the model see all images in training, as the training data maybe of very big sizes, the data in each Epoch is divided into batch sizes. Every epoch has a training and test session, after each Epoch the weights are updated according to the learning rate, optimization algorithms are used to update the learning of a CNN adaptively. When the best weights for training are computed, the model is said to be trained. All the experimental work is carried out in MATLAB simulation tool.