Table 3. Performance comparison of the proposed MSRNet with other models in AD and NC classification based on inter-database cross-validation with the ADNI, AIBL, EDSD, and OASIS databases.

Method	ACC (%)	SEN (%)	SPE (%)	AUC (%)
AGNN	85.2	65.6	92.8	89.7
GAT	84.0	61.9	92.6	88.5
GraphSAGE	87.3	72.9	92.9	91.7
HGNN	77.5	35.8	93.9	72.7
Graph U-Net	77.7	29.5	96.5	71.1
ConvMixer	86.2	78.1	88.8	91.1
3D VGG-16	88.8	79.2	91.4	94.6
3DAN	88.3	73.1	94.2	91.7
Proposed MSRNet	90.6	82.0	93.6	93.9

AD, Alzheimer’s disease; NC, normal control; ACC, accuracy; SEN, sensitivity; SPE, specificity; AUC, area under the curve; MSRNet, multispatial information representation model; ADNI, Alzheimer’s Disease Neuroimaging Initiative database; AIBL, Australian Imaging Biomarkers and Lifestyle; EDSD, European DTI Study on Dementia database; OASIS, Open Access Series of Imaging Studies; DTI, diffusion tensor imaging; AGNN, attention-based graph neural network; GAT, graph attention network; HGNN, hypergraph neural network; 3D VGG-16, 3D visual geometry group 16; 3DAN, 3D attention network.