Probabilistic Principal Component Analysis and Long Short-Term Memory Classifier for Automatic Detection of Alzheimer’s Disease using MRI Brain Images

Abstract

Automatic detection of Alzheimer’s disease using magnetic resonance imaging is a hard task, due to the complexity and variability of the size, location, texture, and shape of the lesions. The objective of this study is to propose a proper feature dimensional reduction and classifier to improve the performance of Alzheimer’s disease detection. At first, the brain images are acquired from Open Access Series of Imaging Studies and National Institute of Mental Health and Neuro Sciences databases. Then, contrast-limited adaptive histogram equalization and normalization technique are applied for improving the visual ability of the collected raw images. Next, discrete wavelet transform is used to transform the denoised images in order to extract the feature vectors, and probabilistic principal component analysis algorithm is developed to decrease the dimension of the extracted features that effectively lessen the “curse of dimensionality” concern. At last, long short-term memory classifier is used for classifying the brain images as Alzheimer’s disease, normal, and mild cognitive impairment. From the simulation result, the proposed system obtained better performance compared with the existing systems and showed 3–11% improvement in recognition accuracy.