Predictive modeling of Alzheimer’s disease progression using multiomics and neuroimaging data with transformer architectures on the ADNI dataset

Dr. AZHAR MuhammadAmjad, AdeenAdeenAmjadSadiq, MuhammadMuhammadSadiqMahmud, Mohammad SultanMohammad SultanMahmudAli, ZeshanZeshanAliHussain, ShafiqShafiqHussain2026-03-312026-03-312025Azhar, M., Amjad, A., Mahmud, M. S., Sadiq, M., Ali, Z., & Hussain S. (2025). Predictive modeling of Alzheimer’s disease progression using multiomics and neuroimaging data with transformer architectures on the ADNI dataset. In IEEE (Ed.). 2025 International conference on frontiers of information technology (FIT). 2025 International Conference on Frontiers of Information Technology (FIT), Islamabad, Pakistan (pp. 1-6). IEEE.http://hdl.handle.net/20.500.11861/27054Alzheimer’s disease is a complex neurodegenerative disorder characterized by progressive cognitive decline. This paper presents the development of advanced predictive models that integrate multiomics data (genomics, proteomics, metabolomics) and neuroimaging from the ADNI dataset by using specialized transformer architectures. Our Efficient Vision Transformer (EViT) for neuroimaging and Efficient TabTransformer (EffiTabTrans) for multiomics data achieved 96.4% diagnostic accuracy, outperforming conventional machine learning baselines in every single metric. External validation on OASIS-3 yielded a strong generalizability of 92.1% accuracy, while optimised architecture reduced training time by 42% for clinical feasibility. Comprehensive ablation studies quantified complementary modality contributions, neuroimaging 58%, multiomics 42%, while SHAP analysis provided insights interpretable for clinical adoption. The framework overcomes major limitations of state-of-the-art AD diagnostics through robust integration of multimodalities and computational efficiency.enNeuroimaging DataMulti-Omics DataTransformer ArchitectureADNI DatasetProteomeExternal ValidationComputational EfficiencyTransformation EfficiencyProgressive Cognitive DeclineReduce Training TimeVision TransformerHealthy ControlsDeep LearningAlzheimer’s DiseaseMild Cognitive ImpairmentMetabolomics DataGenetic Risk ScoreDiagnosis Of Alzheimer’s DiseaseFusion StrategyMultisensory IntegrationHigh Computational EfficiencyPrimary MetricsGPU MemoryPositional EncodingLate FusionEarly FusionPredictive modeling of Alzheimer’s disease progression using multiomics and neuroimaging data with transformer architectures on the ADNI datasetConference Paper10.1109/FIT67061.2025.11333600