Improved Alzheimer Detection using Image Enhancement Techniques and Transfer Learning

Abstract

Alzheimer Disease [AD] is a neuro-degenerative disease which, supported by its ‘progressive nature’ causes loss of function of neurons. Despite substantial research on the application of deep learning algorithms for the identification of Alzheimer's disease, few studies have focused on the image preparation procedures, which are critical in any computer-aided diagnosis system. In fact, the image processing model proposed outlines the criteria for denoising, contrast enhancement, feature extraction, classification, segmentation, and other image processing operations. A central focus of our work is to advance AD detection via transfer learning and image enhancement techniques using magnetic resonance imaging (MRI). Image quality, which is improved during the preprocessing stage, can have a considerable impact on diagnosis. In this research paper, we are representing the image enhancement in multilevel classification of Alzheimer’s disease using the kaggle dataset with a pre-trained VGG16 deep learning architecture. The MRI dataset was enhanced separately with the Contrast Limited Adaptive Histogram Equalization (CLAHE) method, the Fuzzy Color Image Enhancement (FCIE) algorithm, the Hyper column technique, and skull stripping followed by Gaussian smoothing. The retrieved deep features were concatenated to form an ensemble, followed by AD classification with SVM classifier, using all possible combinations of original and enhanced datasets taken three at a time. By combining deep features retrieved from the FC-8 layer of the VGG model trained on original, fuzzy enhanced, and CLAHE datasets, we attained the best accuracy of 99.33%. The obtained results demonstrate that the dataset enhancement strategy has been proven to improve the proposed approach's prediction success.