Angle Classifier for Registration of MRI and CT Brain Images using Deep Learning

Abstract

Image registration in field of medical images is highly recommended for detection brain tumor related diseases. With Deep Learning, features are learnt automatically and it allows the system to quickly iterate complex functions. The paper proposes an image registration methodology for Magnetic Resonance Imaging and Computed Tomography using Deep learning architecture - Convolutional Neural Network. This can identify the orientation of the images. The paper highlights the choice of activation functions for the classifier, trained with 4000 CT and MRI images grouped in 10 classes with angle orientation of 0 - 20 degrees. Experiments indicate the highest accuracy of 95.4 % with clipped Relu activation function, for the proposed architecture trained with 55 epochs. ADAM optimizer provides the highest validation accuracy of 91.28%. A confusion matrix is generated to indicate the classified and misclassified images along with precision and recall values.