Comparative efficacy of histogram-based local descriptors and CNNs in the MRI-based multidimensional feature space for the differential diagnosis of Alzheimer's disease: a computational neuroimaging approach

Abstract

The utilisation of magnetic resonance imaging (MRI) images for the automated detection of Alzheimer's disease has garnered significant attention in recent years. This interest stems from the progress made in machine learning techniques and the possible application of such methods in the field of diagnostics. This study aims to evaluate the performance of 16 histogram-based image texture descriptors and features extracted from 18 pre-trained convolutional neural networks in characterising brain patterns observed in 2D slices of MRI images. The primary objective is to determine the most effective feature types for this task. The characteristics were taken from the magnetic resonance imaging (MRI) dataset given by the Alzheimer's Disease Neuroimaging Initiative (ADNI). The study involved the calculation of features on 2D axial, coronal, and sagittal slices, followed by classification using five binary machine learning algorithms. The objective was to differentiate between individuals with normal cognitive function and those diagnosed with Alzheimer's disease. The proposed methodology additionally facilitated the identification of specific brain areas to be selected for each axis, in order to achieve optimal accuracy. This involved determining the matching feature and classifier combinations.