Alzheimer's Disease Prediction by Spatio-Temporal Feature Fusion for MRI Data

• Published in: 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC) pp. 580 - 585
• Main Authors: S, Harsha Nandhini, J, Aravinth
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.05.2023
• Subjects: 2D-convolution neural network; Alzheimer's disease; Alzheimer's Disease Neuroimaging Initiative; Analytical models; canonical correlation analysis; Correlation; Feature extraction; gray level co-occurrence matrix; hierarchical supervised local CCA; Magnetic resonance imaging; mild cognitive impairment; Neuroimaging; Predictive models; principal component analysis; Radio frequency
• DOI: 10.1109/ICSCCC58608.2023.10176666

Alzheimer's disease (AD) is one of the leading causes of memory loss where patients eventually die. Currently, there is no cure for patients with AD, but patients with Mild Cognitive Impairment (MCI) is the prodromal stage of AD may get diagnosed and have the chance to escape from AD. However, many people with AD remain incurable. Diagnosis of MCI is very important to alleviate complications like brain cell loss and brain shrinking. This can be achieved by developing a spatio-temporal feature fusion aiding early diagnosis of AD. This study proposes a model for spatio-temporal feature fusion using two techniques a) Hierarchical Supervised Local Canonical Correlation Analysis (HSL-CCA) and b) 2D-Convolution Neural Network (2D-CNN). This work uses Magnetic Resonance Imaging (MRI) data from Alzheimer's Disease Neuroimaging Initiative (ADNI) database scans for the 5 stages of AD subjects, and spatial, temporal, and deep features are extracted using Gray-Level Co-occurrence Matrix (GLCM), curvelet transform and 2D-CNN. Principal Component Analysis (PCA) is applied to remove the redundant information, and Short-time Fourier transform (STFT) is used to transform the 1D information into 2D images. Finally, Feature fusion techniques such as: simple concatenation, canonical correlation analysis (CCA), HSL-CCA and 2D-CNN are incorporated to fuse extracted features. The post-processing stage which includes Random Forest (RF) based classification of fused features to classify multiclass of Alzheimer's. The outcome of different feature fusion techniques was compared with several baseline models, the proposed HSL-CCA and 2D CNN fusion approach achieves 84 % and 87% in AD detection.