AI Combines Brain Imaging Features to Identify Cognitive Impairment in CKD
Posted on 19 Nov 2026
Chronic kidney disease (CKD) can affect multiple organs, including the brain, with cognitive impairment emerging as a common complication. Changes in executive function, attention, and memory can reduce quality of life and interfere with treatment adherence, highlighting the need for reliable detection methods. Researchers have developed a deep learning approach that uses multimodal brain MRI to identify cognitive impairment in patients with CKD.
In work associated with Capital Medical University, the research team developed a stage- and cognition-informed neuroimaging feature-selection framework. The framework was designed to identify brain imaging features linked to both CKD progression and cognitive status. These selected features were then used to build a deep learning model called the Dual-Branch Feature-Gated Network (DBFG-Net).
The study included a discovery cohort of 342 patients with CKD. This group comprised 177 patients with cognitive impairment and 165 without cognitive impairment, along with 58 age- and sex-matched healthy controls. The researchers collected multimodal magnetic resonance imaging data using T1-weighted imaging for brain structure, arterial spin labeling for cerebral blood flow, and quantitative susceptibility mapping for tissue susceptibility.
The research process first examined neuroimaging changes across CKD stages 1 to 5. The team also compared imaging features between patients with and without cognitive impairment. Features were retained only when they showed significant stage-related trends and significant differences associated with cognitive status, resulting in 24 selected neuroimaging features that included 17 structural features and seven susceptibility features.
DBFG-Net processed structural and susceptibility data through two separate branches. The model then integrated these complementary imaging inputs through a gated fusion module. This design allowed the system to combine information from different MRI-derived feature groups when identifying cognitive impairment.
The analysis showed that cortical thickness and subcortical volume had significant stage-related trends, with more advanced CKD generally associated with reductions in these structural measures. Tissue susceptibility also changed significantly across CKD stages, although the patterns varied by brain region. Cerebral blood flow did not show a significant linear trend across CKD stages, and it did not significantly differ between patients with and without cognitive impairment.
DBFG-Net achieved an area under the receiver operating characteristic curve of 0.942 in the discovery cohort. Performance was maintained in two independent external validation cohorts, with AUC values of 0.902 and 0.926. Ablation analyses indicated that both structural and susceptibility branches contributed to model performance and suggested that specific cortical and subcortical regions may be relevant to cognitive function in CKD.
The study was limited by its cross-sectional design, which cannot define longitudinal trajectories or establish causal links between kidney dysfunction, brain alterations, and cognitive impairment. Further prospective studies in larger and more diverse populations are needed. The findings suggest that multimodal brain MRI combined with deep learning may provide a technical framework for identifying cognitive impairment in patients with CKD.
Related Links
Capital Medical University