Biomarkers.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging Analysis
• Digital Twin Technology

Background:

• Mixed dementia presents diagnostic and treatment challenges due to its heterogeneity.
• Digital twin technology offers a novel approach for predictive modeling in complex diseases.
• Accurate identification of mixed dementia subtypes is crucial for effective patient management.

Purpose of the Study:

• To construct a digital twin neural marker for mixed dementia using multimodal neuroimage data.
• To enable predictive tasks for delineating the etiologies of mixed dementia.
• To leverage cross-site datasets for robust model development.

Main Methods:

• A Vision Transformer (ViT) model was trained using Masked AutoEncoder (MAE) on T1w and FLAIR neuroimages.
• Federated learning was employed for privacy-preserving, cross-site model training.
• The model was fine-tuned on mixed dementia cohort data to delineate specific etiologies.

Main Results:

• The digital twin neural marker demonstrated reliable performance in differentiating mixed dementia etiologies.
• Latent representations visualized with UMAP showed distinct patterns for different underlying causes.
• The study confirmed the effectiveness of digital twins in analyzing large, multi-site neuroimaging datasets.

Conclusions:

• A digital twin neural marker shows significant potential for addressing the complexities of mixed dementia diagnosis.
• This international collaboration (Korea-UK) developed a powerful tool for mixed dementia research.
• The approach upholds data democracy principles while utilizing extensive datasets.