Alzheimer's Disease Diagnosis With Brain Structural MRI Using Multiview-Slice Attention and 3D Convolution Neural Network

Abstract

Numerous artificial intelligence (AI) based approaches have been proposed for automatic Alzheimer's disease (AD) prediction with brain structural magnetic resonance imaging (sMRI). Previous studies extract features from the whole brain or individual slices separately, ignoring the properties of multi-view slices and feature complementarity. For this reason, we present a novel AD diagnosis model based on the multiview-slice attention and 3D convolution neural network (3D-CNN). Specifically, we begin by extracting the local slice-level characteristic in various dimensions using multiple sub-networks. Then we proposed a slice-level attention mechanism to emphasize specific 2D-slices to exclude the redundancy features. After that, a 3D-CNN was employed to capture the global subject-level structural changes. Finally, all these 2D and 3D features were fused to obtain more discriminative representations. We conduct the experiments on 1,451 subjects from ADNI-1 and ADNI-2 datasets. Experimental results showed the superiority of our model over the state-of-the-art approaches regarding dementia classification. Specifically, our model achieves accuracy values of 91.1 and 80.1% on ADNI-1 for AD diagnosis and mild cognitive impairment (MCI) convention prediction, respectively.

Convolution Neural Networks and Self-Attention Learners for Alzheimer Dementia Diagnosis from Brain MRI

“…Existing MRI-based CAD approaches can use the entire 3D brain volume [ 10 ] or a series of 2D slices extracted from it [ 11 ]. The initial studies relied on traditional algorithmic pipelines (hand-crafted features combined with shallow classifiers) [ 12 ].…”

“…More recently, following the trend in medical imaging [ 13 , 14 , 15 ], Deep Learning (DL) is the most common method for automatic brain feature extraction. Since it depends on many training parameters, using DL on 3D brain volumes significantly increases the computational cost [ 10 ]. In addition, the availability of 3D data is limited, and its use may suffer from the curse of dimensionality [ 16 ], limiting the ability to create accurate models.…”

Multimodal Cross Enhanced Fusion Network for Diagnosis of Alzheimer’s Disease and Subjective Memory Complaints

“…Based on sMRI, Chen et al (11) extracted multi-view slice features and global structural features using multiple slice-level and subject-level subnetworks, which ignored complex spatially information. Liu et al (10) proposed a joint learning multi-task network that performed hippocampus segmentation and then extracted corresponding features for classification.…”

Multimodal cross enhanced fusion network for diagnosis of Alzheimer’s disease and subjective memory complaints

“…In the feature extraction step, existing networks can be classified into 2Dbased, 3D-based, and transformer-based networks according to the architecture. The 2D-based networks perform classification by extracting voxel-level and slice-level features with a small number of network parameters but severe loss of spatial information [10], [11]. The 3D-based networks construct 3D convolutions to extract spatial contextual information natively by applying original images.…”

“…Based on sMRI, Chen et al [11] extracted multi-view slice features and global structural features by multiple slicelevel and subject-level subnetworks which ignored spatially complex information. Liu et al [10] proposed a joint learning multi-task network that performed segmentation of the hippocampus and then extracted the corresponding features for classification.…”