Flowchart of the study. (IMAGE)
Caption
The workflow consists of 2 major steps. The blue panel depicts the construction of a diffusion tensor imaging (DTI)-based DenseNet model to identify subcortical vascular cognitive impairment (SVCI) from subcortical ischemic vascular disease (SIVD) patients using diffusion scalar image combinations. Diffusion scalar images were derived from preprocessed DTI images. An unsupervised domain adaptation strategy was applied to enhance the model’s performance on unseen target-domain data, and model performance was evaluated on both internal and target-domain test sets. The neuropsychological relevance of model outputs—including SVCI probability and salient white matter regions—was further assessed. The orange panel illustrates the individual-level cognitive profiling based on images. Voxel-wise mutual information (MI) maps were computed between diffusion scalar images and 6 neuropsychological scales to identify domain-specific white matter correlates, and structural similarity index measure (SSIM) was calculated between individual-level model-derived salient weight maps and MI maps. Unsupervised clustering of SSIM scores enabled imaging-based cognitive risk stratification across cognitive domains.
Credit
Yi Tang, Department of Neurology & Innovation Center for Neurological Disorders, Xuanwu Hospital, Capital Medical University, National Center for Neurological Disorders.
Usage Restrictions
News organizations may use or redistribute this image, with proper attribution, as part of news coverage of this paper only.
License
Licensed content