Share：
Share this content in WeChat
X
[Chinese] [PDF] 53 1
• Clinical Article •
Application value of an interpretable 3D directional attention network based on prior knowledge of directional atrophy in Alzheimer's disease diagnosis
WANG Zihao  ZHOU Jinliang  ZHENG Qingqing  YI Wei  XIAO Jinyu  REN Rui 

DOI:10.12015/issn.1674-8034.2026.05.007.


[Abstract] Objective To develop a clinically inspired 3D directional attention deep learning model for the automatic classification of Alzheimer's disease (AD) and cognitively normal (CN) individuals, and to evaluate its generalization ability in an independent real-world clinical setting as well as its consistency with diagnoses made by radiologists of varying experience levels.Materials and Methods A total of 621 subjects (275 AD, 346 CN) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were retrospectively included as the development set, which was randomly split into a training set (n = 496) and an internal validation set (n = 125) at an 8∶2 ratio. Additionally, 90 participants (60 AD, 30 CN) from a local hospital cohort were included as an independent external test set. A 3D directional attention module (DirectionTripleAttention3D), comprising spatial, channel, and Sobel operator-inspired directional attention branches, was integrated into a 3D DenseNet-121 backbone to capture morphological gradient features of cortical atrophy. Three-dimensional gradient-weighted class activation mapping (3D Grad-CAM) was employed to visualize the model's regions of interest. The diagnostic performance of the model was compared with that of two radiologists with different levels of experience (junior and senior).Results In the ADNI internal validation set, the DenseNet121-DirAtt3D model achieved an area under the curve (AUC) of 0.924 (95% CI: 0.872 to 0.967) and an accuracy of 88.0%. In the independent external test set from our institution, the model yielded an AUC of 0.906 (95% CI: 0.818 to 0.976) and an accuracy of 88.9%, demonstrating good robustness. Human–machine comparison showed that, compared with the junior radiologist, the model exhibited a significant advantage in overall classification performance (McNemar test, χ2 = 7.314, adjusted P = 0.014). Meanwhile, the overall diagnostic performance of the model was comparable to that of the senior radiologist, with no significant difference in AUC between the two (DeLong test, Z = 0.40, adjusted P = 0.691). The regions highlighted by the model were mainly located in the bilateral hippocampus, medial temporal lobe, and parahippocampal gyrus, and were visually consistent with the atrophy pattern of AD.Conclusion The 3D directional attention-based deep learning model demonstrates excellent diagnostic performance and generalization capabilities in the classification of AD. Its decision-making basis possesses anatomical interpretability. Consequently, it holds promise as an auxiliary decision-support tool to enhance standardization and consistency in the imaging diagnosis of AD.
[Keywords] Alzheimer disease；magnetic resonance imaging；deep learning；directional attention；human-machine comparison

WANG Zihao1   ZHOU Jinliang1   ZHENG Qingqing2   YI Wei1   XIAO Jinyu1   REN Rui1*  

1 Department of Radiology, Binzhou Medical University Affiliated Hospital, Binzhou 256603, China

2 Supervision Section Ⅱ, Binzhou Center for Disease Control and Prevention, Binzhou 256600, China

Corresponding author: REN R, E-mail: 13954344825@163.com

Conflicts of interest   None.

Received  2026-01-05
Accepted  2026-05-08
DOI: 10.12015/issn.1674-8034.2026.05.007
DOI:10.12015/issn.1674-8034.2026.05.007.

[1]
ZHI N, REN R, QI J, et al. The China Alzheimer report 2025[J/OL]. Gen Psychiatr, 2025, 38(4): e102020 [2026-01-05]. https://doi.org/10.1136/gpsych-2024-102020. DOI: 10.1136/gpsych-2024-102020.
[2]
JIA L, DU Y, CHU L, et al. Prevalence, risk factors, and management of dementia and mild cognitive impairment in adults aged 60 years or older in China: A cross-sectional study[J/OL]. Lancet Public Health, 2020, 5(12): e661-e671 [2026-01-05]. https://doi.org/10.1016/S2468-2667(20)30185-7. DOI: 10.1016/S2468-2667(20)30185-7.
[3]
Alzheimer's Association. 2025 Alzheimer's disease facts and figures[J/OL]. Alzheimers Dement, 2025, 21(4): e70235 [2026-01-05]. https://doi.org/10.1002/alz.70235. DOI: 10.1002/alz.70235.
[4]
FRISONI G B, FOX N C, JACK C R, et al. The clinical use of structural MRI in Alzheimer disease[J]. Nat Rev Neurol, 2010, 6(2): 67-77. DOI: 10.1038/nrneurol.2009.215.
[5]
ZHOU Y, ZHOU Q D, YIN C R. Research progress of PET/MR in Alzheimer's disease under ATN diagnostic framework[J]. Chin J Magn Reson Imaging, 2024, 15(6): 159-165. DOI: 10.12015/issn.1674-8034.2024.06.025.
[6]
ZHOU J, WEI Y, LI X, et al. A deep learning model for early diagnosis of Alzheimer's disease combined with 3D CNN and video swin transformer[J/OL]. Sci Rep, 2025, 15(1): 23311 [2026-01-05]. https://doi.org/10.1038/s41598-025-05568-y. DOI: 10.1038/s41598-025-05568-y.
[7]
MOUSAVI S M, MOULAEI K, AHMADIAN L. Classifying and diagnosing Alzheimer's disease with deep learning using 6735 brain MRI images[J/OL]. Sci Rep, 2025, 15(1): 22721 [2026-01-05]. https://doi.org/10.1038/s41598-025-08092-1. DOI: 10.1038/s41598-025-08092-1.
[8]
KIM J S, HAN J W, BAE J B, et al. Deep learning-based diagnosis of Alzheimer's disease using brain magnetic resonance images: An empirical study[J/OL]. Sci Rep, 2022, 12(1): 18007 [2026-01-05]. https://doi.org/10.1038/s41598-022-22917-3. DOI: 10.1038/s41598-022-22917-3.
[9]
ZHANG H, ZHANG S, SUN J W, et al. Research progress of deep learning in MRI image classification of brain tumors[J]. Chin J Magn Reson Imaging, 2023, 14(1): 166-171, 193. DOI: 10.12015/issn.1674-8034.2023.01.031.
[10]
XU B H, DING C, XU G Z. Research on application of convolutional neural networks in diagnosis of Alzheimer's disease[J]. J Biomed Eng, 2021, 38(1): 169-177, 184. DOI: 10.7507/1001-5515.202005018.
[11]
ZHAO F, WU Z, LI G. Deep learning in cortical surface-based neuroimage analysis: A systematic review[J]. Intell Med, 2023, 3(1): 46-58. DOI: 10.1016/j.imed.2022.06.002.
[12]
LU B, LI H X, CHANG Z K, et al. A practical Alzheimer's disease classifier via brain imaging-based deep learning on 85,721 samples[J/OL]. J Big Data, 2022, 9(1): 101 [2026-01-05]. https://doi.org/10.1186/s40537-022-00650-y. DOI: 10.1186/s40537-022-00650-y.
[13]
ISLAM J, FURQON E N, FARADY I, et al. Enhancing the reliability of Alzheimer's disease prediction in MRI images[J/OL]. Comput Biol Med, 2025, 197: 111111 [2026-01-05]. https://doi.org/10.1016/j.compbiomed.2025.111111. DOI: 10.1016/j.compbiomed.2025.111111.
[14]
VAROQUAUX G, CHEPLYGINA V. Machine learning for medical imaging: methodological failures and recommendations for the future[J/OL]. NPJ Digit Med, 2022, 5(1): 48 [2026-01-05]. https://doi.org/10.1038/s41746-022-00592-y. DOI: 10.1038/s41746-022-00592-y.
[15]
THOMPSON P M, HAYASHI K M, de ZUBICARAY G, et al. Dynamics of gray matter loss in Alzheimer's disease[J]. J Neurosci, 2003, 23(3): 994-1005. DOI: 10.1523/JNEUROSCI.23-03-00994.2003.
[16]
WEINER M W, VEITCH D P, AISEN P S, et al. The Alzheimer's disease neuroimaging initiative: A review of papers published since its inception[J]. Alzheimers Dement, 2012, 8(1Suppl): S1-S68. DOI: 10.1016/j.jalz.2011.09.172.
[17]
MCKHANN G M, KNOPMAN D S, CHERTKOW H, et al. The diagnosis of dementia due to Alzheimer's disease: Recommendations from the national institute on aging-Alzheimer's association workgroups on diagnostic guidelines for Alzheimer's disease[J]. Alzheimers Dement, 2011, 7(3): 263-269. DOI: 10.1016/j.jalz.2011.03.005.
[18]
SCHELTENS P, LEYS D, BARKHOF F, et al. Atrophy of medial temporal lobes on MRI in "probable" Alzheimer's disease and normal ageing: diagnostic value and neuropsychological correlates[J]. J Neurol Neurosurg Psychiatry, 1992, 55(10): 967-972. DOI: 10.1136/jnnp.55.10.967.
[19]
PASQUIER F, LEYS D, WEERTS J G, et al. Inter- and intraobserver reproducibility of cerebral atrophy assessment on MRI scans with hemispheric infarcts[J]. Eur Neurol, 1996, 36(5): 268-272. DOI: 10.1159/000117270.
[20]
KOEDAM E L G E, LEHMANN M, VAN DER FLIER W M, et al. Visual assessment of posterior atrophy development of a MRI rating scale[J]. Eur Radiol, 2011, 21(12): 2618-2625. DOI: 10.1007/s00330-011-2205-4.
[21]
KOO T K, LI M Y. A guideline of selecting and reporting intraclass correlation coefficients for reliability research[J]. J Chiropr Med, 2016, 15(2): 155-163. DOI: 10.1016/j.jcm.2016.02.012.
[22]
ÁVILA-JIMÉNEZ J L, CANTÓN-HABAS V, CARRERA-GONZÁLEZ M del P, et al. A deep learning model for Alzheimer's disease diagnosis based on patient clinical records[J/OL]. Comput Biol Med, 2024, 169: 107814 [2026-01-05]. https://doi.org/10.1016/j.compbiomed.2023.107814. DOI: 10.1016/j.compbiomed.2023.107814.
[23]
HAN Y M, LI Y J, ZHANG H, et al. Application value of deep learning in imaging research of Alzheimer's disease conversion prediction[J]. J Pract Med, 2025, 41(9): 1413-1424. DOI: 10.3969/j.issn.1006-5725.2025.09.021.
[24]
GHOSH T, PALASH M I A, YOUSUF M A, et al. A robust distributed deep learning approach to detect Alzheimer's disease from MRI images[J/OL]. Mathematics, 2023, 11(12): 2633 [2026-01-05]. https://doi.org/10.3390/math11122633. DOI: 10.3390/math11122633.
[25]
RAHMAN S, RAHMAN M, BHATT S, et al. NeuroNet-AD: A multimodal deep learning framework for multiclass Alzheimer's disease diagnosis[J/OL]. Bioengineering, 2025, 12: 1107 [2026-01-05]. https://doi.org/10.3390/bioengineering12101107. DOI: 10.3390/bioengineering12101107.
[26]
ANSARI A S, MOHAMMADI M S, CATTANI C, et al. An advanced multimodal image fusion model for accurate detection of Alzheimer's disease using MRI and PET[J/OL]. Front Med Technol, 2025, 7: 1699821 [2026-01-05]. https://doi.org/10.3389/fmedt.2025.1699821. DOI: 10.3389/fmedt.2025.1699821.
[27]
FISCHL B, DALE A M. Measuring the thickness of the human cerebral cortex from magnetic resonance images[J]. Proc Natl Acad Sci U S A, 2000, 97(20): 11050-11055. DOI: 10.1073/pnas.200033797.
[28]
CHAN S C K, SHI L, HUANG B, et al. Directional adaptive shuffle-based visual state-space models for medical image restoration[C]. GEE JC, ALEXANDERD C, HONG J, et al., eds.//Medical Image Computing and Computer Assisted Intervention-MICCAI 2025. Cham: Springer Nature Switzerland, 2025: 160-170. DOI: 10.1007/978-3-031-72069-7_16.
[29]
GUAN H, LIU M. Domain adaptation for medical image analysis: a survey[J]. IEEE Trans Biomed Eng, 2022, 69(3): 1173-1185. DOI: 10.1109/TBME.2021.3117407.
[30]
BRON E E, KLEIN S, PAPMA J M, et al. Cross-cohort generalizability of deep and conventional machine learning for MRI-based diagnosis and prediction of Alzheimer's disease[J/OL]. Neuroimage Clin, 2021, 31: 102712 [2026-01-05]. https://doi.org/10.1016/j.nicl.2021.102712. DOI: 10.1016/j.nicl.2021.102712.
[31]
WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]. FERRARIV, HEBERTM, SMINCHISESCUC, et al., eds.//Computer Vision-ECCV 2018. Cham: Springer International Publishing, 2018: 3-19. DOI: 10.1007/978-3-030-01234-2_1.
[32]
HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C/OL]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2018: 7132-7141 [2025-12-18]. https://ieeexplore.ieee.org/document/8578843. DOI: 10.1109/CVPR.2018.00745.
[33]
KANOPOULOS N, VASANTHAVADA N, BAKER R L. Design of an image edge detection filter using the sobel operator[J]. IEEE J Solid-state Circuits, 1988, 23(2): 358-367. DOI: 10.1109/4.996.
[34]
ZHANG L, WANG X, YANG D, et al. Generalizing deep learning for medical image segmentation to unseen domains via deep stacked transformation[J]. IEEE Trans Med Imaging, 2020, 39(7): 2531-2540. DOI: 10.1109/TMI.2020.2973595.
[35]
WEN J, THIBEAU-SUTRE E, DIAZ-MELO M, et al. Convolutional neural networks for classification of Alzheimer's disease: Overview and reproducible evaluation[J/OL]. Med Image Anal, 2020, 63: 101694 [2026-01-05]. https://doi.org/10.1016/j.media.2020.101694. DOI: 10.1016/j.media.2020.101694.
[36]
MARKUS A F, KORS J A, RIJNBEEK P R. The role of explainability in creating trustworthy artificial intelligence for health care: A comprehensive survey of the terminology, design choices, and evaluation strategies[J/OL]. J Biomed Inf, 2021, 113: 103655 [2026-01-05]. https://doi.org/10.1016/j.jbi.2020.103655. DOI: 10.1016/j.jbi.2020.103655.
[37]
JACK C R, BENNETT D A, BLENNOW K, et al. NIA-AA research framework: Toward a biological definition of Alzheimer's disease[J]. Alzheimers Dement, 2018, 14(4): 535-562. DOI: 10.1016/j.jalz.2018.02.018.
[38]
BRAAK H, BRAAK E. Neuropathological stageing of Alzheimer-related changes[J]. Acta Neuropathol, 1991, 82(4): 239-259. DOI: 10.1007/BF00308809.

PREV Investigating bidirectional differences in glymphatic system function between vestibular migraine and migraine without aura using DTI-ALPS
NEXT Brain functional impairment in drug-free patients with type 2 diabetes: A resting-state fMRI study
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn