MRI Brain Image Classification Using Convolutional Neural Networks And Transfer Learning

Abstract

Alzheimer's disease (AD) is a neurodegenerative disorder. There is no particular cure for Alzheimer's disease. Accurate and early diagnosis of AD could assist patients in receiving appropriate care. However, diagnosing AD in brain MRI images is a difficult task that depends on the presence of experienced radiologists or medical professionals. As one MRI exam might generate thousands of images, it typically takes several weeks for the results to be obtained. Many researchers use statistical and machine learning methods to diagnose Alzheimer's disease. Deep Learning algorithms have demonstrated human-level competence in a variety of fields. Deep learning, especially convolutional neural networks (CNN), is becoming popular because of its state-of-the-art performance in many computer vision tasks such as visual object classification, object detection, and segmentation. Transfer learning is a technique that can be used with CNNs to improve their performance. The purpose of this project is to develop a model of brain MRI image classification for Alzheimer's disease diagnosis by using CNN and transfer learning. In this project, the modified VGG16 model with fine-tuning was proposed, and MRI data from the OASIS database was used to classify Alzheimer's disease into three (3) different classes, which are AD, MCI and NC. The model is developed using Google Collaboratory and Adam's optimization algorithm. The proposed model has achieved a training accuracy of 98.56% and the validation accuracy of 90.24%.