Medical Image Classification using Deep Learning

Abstract

Sometimes triggering cell multiplication to form tissues goes wrong and ends up creating tumor. Early detection and classifying it to malignant(cancer-causing) or benign(not cancer causing) at initial stages of the same is important in order to prevent its further growth. MRI (Magnetic Resonance Imaging) is a medical imaging technique used by radiologists to study and analyze medical images. Doing critical analysis like this not only can create unnecessary delay but also the accuracy for the same could be very less due to human errors. The main objective of this project is to apply machine learning techniques to make systems capable enough to perform such critical analysis faster having higher accuracy and efficiency levels. The implementation is done with the use of AlexNet, ZFNet and a proposed architecture having the combination of convolution, pooling and fully connected layers that have been used for classification of brain tumor into two classes, tumorous and non-tumorous. The proposed deep learning convolution neural network architecture is classifying brain tumor in two stages. In the first stage, in tumorous or non-tumorous and then the predicted tumorous images are further classified into Astrocytoma, Glioblastoma Multiforme (GBM), Mixed Glioma and Oligodendroglioma.The classification accuracy achieved by AlexNet, ZFNet and the proposed architecture are 63.56%, 84.42%, and 95.41% respectively, signifying that the proposed architecture is much better than the existing architectures irrespective of their training accuracy of 93.18%, 96.61%, 91.67% respectively. The data set REMBRANDT was used for the implementation of this project.