Convolutional Neural Networks (CNNs) have recently been introduced to Alzheimer's Disease (AD) diagnosis. Despite their encouraging prospects, most of the existing models only process AD-related brain atrophy on a single spatial scale, and have high computational complexity. Here, we propose a novel Attention-based 3D Multi-scale CNN model (AMSNet), which can better capture and integrate multiple spatial-scale features of AD, with a concise structure. For the binary classification between 384 AD patients and 389 Cognitively Normal (CN) controls using sMRI scannings, AMSNet achieves remarkable overall performance (91.3% accuracy, 88.3% sensitivity, and 94.2% specificity) with fewer parameters and lower computational load, generally surpassing seven comparative models. Furthermore, AMSNet generalizes well in other AD-related classification tasks, such as the three-way classification (AD-MCI-CN). Our results manifest the feasibility and efficiency of the proposed multi-scale spatial feature integration and attention mechanism used in AMSNet for AD classification, and provide potential biomarkers to explore the neuropathological causes of AD.

Background: Alzheimer’s disease (AD) is a common neurodegenerative disease occurring in the elderly population. The effective and accurate classification of AD symptoms by using functional magnetic resonance imaging (fMRI) has a great significance for the clinical diagnosis and prediction of AD patients. Methods: Therefore, this paper proposes a new method for identifying AD patients fromhealthy subjects by using functional connectivities (FCs) between the activity voxels in the brain based on fMRI data analysis. Firstly, independent component analysis is used to detect the activity voxels in the fMRI signals of AD patients and healthy subjects; Secondly, the FCs between the common activity voxels of the two groups are calculated, and then the FCs with significant differences are further identified by statistical analysis between them; Finally, the classification of AD patients fromhealthy subjects is realized by using FCs with significant differences as the feature samples in support vector machine. Results: The results show that the proposed identification method can obtain higher classification accuracy, and the FCs between activity voxels within prefrontal lobe as well as those between prefrontal and parietal lobes play an important role in the prediction of AD patients. Furthermore, we also find that more brain regions and much more voxels in some regions are activity in AD group compared with health control group. Conclusion: It has a great potential value for the AD pathogenesis mechanism study.

Graph deep learning (GDL) has demonstrated impressive performance in predicting population-based brain disorders (BDs) through the integration of both imaging and non-imaging data. However, the effectiveness of GDL-based methods heavily depends on the quality of modeling multi-modal population graphs and tends to degrade as the graph scale increases. Moreover, these methods often limit interactions between imaging and non-imaging data to node-edge interactions within the graph, overlooking complex inter-modal correlations and resulting in suboptimal outcomes. To address these challenges, we propose MM-GTUNets, an end-to-end Graph Transformer-based multi-modal graph deep learning (MMGDL) framework designed for large-scale brain disorders prediction. To effectively utilize rich multi-modal disease-related information, we introduce Modality Reward Representation Learning (MRRL), which dynamically constructs population graphs using an Affinity Metric Reward System (AMRS). We also employ a variational autoencoder to reconstruct latent representations of non-imaging features aligned with imaging features. Based on this, we introduce Adaptive Cross-Modal Graph Learning (ACMGL), which captures critical modality-specific and modality-shared features through a unified GTUNet encoder, taking advantages of Graph UNet and Graph Transformer, along with a feature fusion module. We validated our method on two public multi-modal datasets ABIDE and ADHD-200, demonstrating its superior performance in diagnosing BDs. Our code is available at https://github.com/NZWANG/MM-GTUNets.

Compared with independent component analysis (ICA), constrained ICA (CICA) has unique advantages in functional magnetic resonance image (fMRI) data analysis by introducing some priori information into the estimation process. However, there are still some controversies in the current CICA methods, such as how to choose the threshold parameter to restrain the similarity, and how to reduce the accuracy requirements for the priori information. In this paper, we propose a new constrained spatiotemporal ICA (CSTICA) method based on the framework of multi-objective optimization, where the inequality constraint of traditional CICA method is transformed into the objective optimization function of CSTICA, and both temporal and spatial priori information are included simultaneously. The simulated, hybrid and real fMRI data experiments are designed to evaluate the performance of the proposed CSTICA method in comparison with classical ICA and CICA methods. Compared with the traditional CICA methods, CSTICA has circumvented the problem of threshold parameter selection. Furthermore, the experimental results demonstrate that the source recovery ability of CSTICA has been improved to a certain extent especially in the cases of priori information with low accuracies. Meanwhile, the results also indicate that CSTICA reduces dependency on the accuracy of priori information.