Table 2 Comparison of different models for AD vs. CN classification, trained on the ADNI training dataset and tested on the ADNI test dataset

Only using 3D images

3D Trans-ResNet [20]

3D CNN + 2D Transformer

0.9143

0.8431

0.9815

0.9053

0.9683

3D ResNet [21]

3D CNN

0.8095

0.7843

0.8333

0.8000

0.9187

3D Swin Transformer [22]

3D Transformer

0.8857

0.8431

0.9258

0.8776

0.9330

Only using 2D slices

DE-ViT [23]

2D Transformer

0.9048

0.9412

0.9704

0.9057

0.9563

2D ResNet [24]

2D CNN

0.7905

0.8431

0.7407

0.7963

0.8228

3D images + 2D slices

M3T [25]

3D CNN + 2D CNN + 2D Transformer

0.9616

0.9412

0.9815

0.9600

0.9899

MHAGuideNet (ours)

3D CNN + 2D CNN + 2D Transformer

0.9758

0.8863

0.9989

0.9398

0.9931