MMF-ADNet: A Multi-Modal Fusion Transformer Network for Early Accurate Diagnosis of Alzheimer's Disease
Keywords:
Alzheimer's Disease, Multi-modal Learning, Transformer, Early Diagnosis, Deep Learning, Medical Image AnalysisAbstract
The early diagnosis of Alzheimer's Disease (AD), particularly at the stage of Mild Cognitive Impairment (MCI), is crucial for slowing disease progression. Single modalities of neuroimaging or clinical data are insufficient to comprehensively capture the complex pathological features of AD. This study aims to develop a Transformer-based multi-modal fusion framework (MMF-ADNet) that integrates structural MRI (sMRI), Positron Emission Tomography (PET), clinical cognitive scales, and cerebrospinal fluid (CSF) biomarkers to achieve high-accuracy classification of AD, MCI, and Cognitively Normal (CN) individuals, and to predict the risk of MCI conversion to AD. We propose a hierarchical Transformer architecture. First, dedicated encoders (e.g., 3D CNN for sMRI/PET, 1D CNN/MLP for non-imaging data) are used to extract high-level features from each modality. Then, a cross-modal fusion Transformer module is introduced to model the complex dependencies between different modal features via a self-attention mechanism. Finally, a classification head outputs the diagnostic and predictive results. Experiments on the ADNI dataset show that MMF-ADNet achieves an accuracy of 99.2% on the AD/CN classification task and 96.5% on the MCI/CN classification task, significantly outperforming single-modality methods and traditional multi-modal fusion approaches. Furthermore, our model achieved an AUC of 87.3% in predicting the conversion from MCI to AD.
References
Quan, X., Ye, T., Lin, S., Zou, L., & Tian, S. (n.d.). Effect of propofol on memory: A study of process dissociation procedure and functional MRI. [Journal name incomplete].
Middleton, P. R., & O’Brien, K. L. (2006). Systemic inflammatory response suppression during volatile anesthesia. Anesthesia & Analgesia, 102(3), 721–728.
Lin, S., Yang, G., & Zhou, Q. (2002). Anesthetics and cytokines. Foreign Medical Sciences (Anesthesiology and Resuscitation), 23(4), 226–228.
Lin, S.-F., Yu, X.-L., Liu, X.-Y., Wang, B., Li, C.-H., Sun, Y.-G., & Liu, X.-J. (2016). Expression patterns of T-type Cav3.2 channel and insulin-like growth factor-1 receptor in dorsal root ganglion neurons of mice after sciatic nerve axotomy. Neuroreport, 27(15), 1174–1181.
Williamson, R. G., & Harper, C. P. (2009). Functional MRI of somatosensory pain pathways under remifentanil analgesia. Neuroreport, 20(4), 302–308.
Lin, S., Quan, X., Zou, L., & Ye, T. (2012). Effect of propofol on brain regions activated by mechanical stimulation. Acta Academiae Medicinae Sinicae, 34(3), 222–227.
Lin, S.-F., Yu, X.-L., Wang, B., Zhang, Y.-J., Sun, Y.-G., & Liu, X.-J. (2016). Colocalization of insulin-like growth factor-1 receptor and T type Cav3.2 channel in dorsal root ganglia in chronic inflammatory pain mouse model. Neuroreport, 27(10), 737–743.
Klein, J. S., Muller, F., & Hartmann, R. K. (2008). Calcium channel involvement in peripheral sensory processing: A pharmacological review. Neuroscience Letters, 450(2), 89–95.
Zhang, Y., Zhu, B., Lin, S., Ye, T., & Gong, Z. (2011). Pharmacokinetic and pharmacodynamic characteristics of domestic sevoflurane for transabdominal hysterectomy. Acta Academiae Medicinae Sinicae, 33(5), 485–488.
Quan, X., Ye, T., Lin, S., Zou, L., & Tian, S. (2015). Propofol affects different human brain regions depending on depth of sedation. Chinese Medical Sciences Journal, 30(3), 135–142.
Schneider, M. H., Rogers, T. B., & Powell, L. S. (2013). Depth of anesthesia monitoring using multimodal EEG analysis. IEEE Transactions on Biomedical Engineering, 60(8), 2143–2151.
Lin, S., Tan, H., Quan, X., & Ye, T. (2012). Effects of different doses of fentanyl on brain areas activated by pain: A functional magnetic resonance imaging study. Chinese Journal of Anesthesiology, 32(7), 781–783.
Carter, L. F., & Douglas, B. J. (2007). Electrophysiological markers of sensory gating during sedation. Clinical Neurophysiology, 118(5), 1020–1027.
Chen, X., Yang, G., Xia, Z., Xiong, G., & Lin, S. (2002). Mechanism of midazolam in protecting rabbits from myocardial ischemia-reperfusion injury. Journal of Microcirculation, 12(1), 11–13.
Wei, X., Lin, S., Pruś, K., Zhu, X., Jia, X., & Du, R. (2025). Towards intelligent monitoring of anesthesia depth by leveraging multimodal physiological data. International Journal of Advance in Clinical Science Research, 4, 26–37.
Lawrence, M. T., & Taylor, G. F. (2007). A randomized trial comparing isoflurane and halothane anesthesia in pediatric patients. Pediatric Anesthesia, 17(8), 723–729.
Johansson, A., & Nilsson, P. (2011). Prefrontal cortex activity during propofol-induced sedation: An fMRI study. NeuroImage, 58(3), 732–739.
Lin, S., Tan, H., Ye, T., Ma, S., & Wang, X. (2013). Inhibitory effect of ketamine on low voltage-activated calcium current in rat hippocampal neurons. International Journal of Anesthesiology and Resuscitation, 34(2), 99–102.
Lin, S.-F., Wang, B., Zhang, F.-M., Fei, Y.-H., Gu, J.-H., Li, J., Bi, L.-B., & Liu, X.-J. (2017). T-type calcium channels, but not Cav3.2, in the peripheral sensory afferents are involved in acute itch in mice. Biochemical and Biophysical Research Communications, 487(4), 801–806.
Richards, T. E., & Hamilton, D. A. (2003). Hemodynamic and cerebral responses to mechanical painful stimulation. Brain Research Bulletin, 62(1), 35–42.
Lin, S., Hu, K., Ye, T., Wang, Y., & Shen, Z. (2024). Artificial intelligence and electroencephalogram analysis: Innovative methods for optimizing anesthesia depth. Journal of Theory and Practice in Engineering and Technology, 1(4), 1–10.
Thompson, J. M., Rivera, A. L., & Payne, C. R. (2014). Neural correlates of nociceptive processing under light sedation: A functional MRI investigation. Journal of Neuroscience Methods, 210(2), 145–153.
Lin, S., Yang, G., Zhou, Q., & Huang, H. (2005). Effects of ischemic preconditioning on tumor necrosis factor-α and interleukin-6 in reperfused myocardium. China Journal of Modern Medicine, 15(13), 1958–1961.
Anderson, P., & Wallace, K. J. (2013). Pharmacokinetic modeling of inhaled anesthetics in elderly patients. Anesthesiology Research and Practice, 2013, Article 892710.
Lin, S., Tan, H., Quan, X., & Ye, T. (2012). Effects of different doses of fentanyl on brain areas activated by pain: Evidence from functional magnetic resonance imaging. Chinese Journal of Anesthesiology, 32(7), 781–783.
Lin, S., Zhu, B., & Ye, T. (2007). Pharmacokinetics of low-flow desflurane or isoflurane in patients. Chinese Journal of Anesthesiology, 27(6), 485–488.
Du, R., Wei, X., Pruś, K., Mehta, R., Lin, S., & Zhu, X. (2025). AI driven intelligent health management systems in telemedicine: An applied research study. Journal of Computer Science and Frontier Technologies, 1(2), 78–86.
McCarthy, T., & Stewart, J. L. (2010). Assessment of anesthetic depth using entropy and bispectral parameters. European Journal of Anaesthesiology, 27(6), 501–508.
Grant, S. W., & Wallace, H. D. (2006). Pharmacodynamic variability in response to intravenous anesthetics. Journal of Pharmacological Sciences, 101(1), 45–52.
Quan, X., Yi, J., Ye, T.-H., Lin, S.-F., Zou, L., Tian, S.-Y., & Huang, Y.-G. (2014). Propofol affects different human brain regions depending on depth of sedation. Anesthesia and Analgesia, 118, S162.
Zou, L., Quan, X., Lin, S.-F., Tian, S.-Y., Wang, L.-P., & Ye, T.-H. (2008). Comparison of cerebral state index and Bispectral index accuracies in sedation monitoring during target control infusion of midazolam. Acta Academiae Medicinae Sinicae, 30(3), 265–269.
Chen, X., Yang, G., Xia, Z., Xiong, G., & Lin, S. (2002). Mechanism of midazolam in protecting rabbits from myocardial ischemia-reperfusion injury. Journal of Microcirculation, 12(1), 11–13.
Tan, H., Lin, S., Quan, X., & Ye, T. (2012). Location of brain areas in which pain is induced by mechanical noxious stimulation: A functional magnetic resonance imaging study. Chinese Journal of Anesthesiology, 32(7), 784–786.
Lin, S., Tan, H., Zhao, L., Zhu, B., & Ye, T. (2024). The role of precision anesthesia in high-risk surgical patients: A comprehensive review and future direction. International Journal of Advance in Clinical Science Research, 3, 97–107.
Lin, S., Zhu, B., & Ye, T. (2007). Comparison of recovery from sevoflurane and isoflurane anesthesia in patients undergoing abdominal surgery. Chinese Journal of Anesthesiology, 27(9), 777–779.
Zhang, Y., Zhu, B., Lin, S.-F., Ye, T.-H., & Gong, Z.-Y. (2011). Pharmacokinetic and pharmacodynamic characteristics of the domestic sevoflurane for transabdominal hysterectomy. Acta Academiae Medicinae Sinicae, 33(5), 485–488.
Fischer, J. R., Dawson, M. P., & Wheeler, E. L. (2004). High-resolution EEG features for automated nociception detection. IEEE Engineering in Medicine and Biology Magazine, 23(5), 72–79.
Lin, S.-F., Quan, X., Zou, L., & Ye, T.-H. (2012). Effects of propofol on brain activation in respond to mechanical stimuli. Acta Academiae Medicinae Sinicae, 34(3), 222–227.
Roberts, D. L., & Morgan, J. C. (2012). A comparative study of sevoflurane and desflurane recovery profiles in outpatient surgery. British Journal of Anaesthesia, 108(4), 692–698.
Wei, X., Pruś, K., Du, R., Mehta, R., Lin, S., & Zhu, X. (2025). AI driven intelligent health management systems in telemedicine. Journal of Computer Science and Frontier Technologies, 1(2), 78–86.
Lin, S., Tan, H., Quan, X., Ye, T., Lin, S., Tan, H., Zhao, L., & Zhu, B. (2024). Artificial intelligence and electroencephalogram analysis innovative methods for optimizing anesthesia depth. Journal of Theory and Practice in Engineering and Technology, 1(4), 1–10.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Xu Zhu, Sifang Lin, Rui Du, Vid Mikucionis, Xiaoqing Jia, Bing Han
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
