Deep learning framework improves vascular cognitive impairment diagnosis

Subcortical ischemic vascular disease (SIVD), driven by cerebral small vessel disease, is commonly characterized by white matter hyperintensities and multiple lacunar infarcts, and a substantial proportion of patients gradually develop cognitive decline and progress to subcortical vascular cognitive impairment (SVCI). Early differentiation of SVCI from SIVD without cognitive impairment is important for slowing cognitive decline and guiding intervention strategies. However, current SVCI diagnosis usually depends on combined assessment of clinical symptoms, structural MRI, and neuropsychological scales, which can be time-consuming and vulnerable to assessment conditions and subjective bias, especially in elderly or resource-limited settings. In addition, white matter hyperintensities on conventional structural MRI are also common in older adults, limiting their specificity and sensitivity for early microstructural white matter injury.

Diffusion tensor imaging (DTI) can provide more sensitive information about white matter microstructure, while deep learning has the potential to automatically extract disease-relevant imaging features. Therefore, using DTI and interpretable deep learning to accurately identify SVCI and further profile individualized risks across cognitive domains has become an important direction for advancing precise diagnosis and personalized intervention in SVCI."

Miao He, author, researcher, Capital Medical University

This study developed a diffusion tensor imaging (DTI)-based deep learning framework to identify subcortical vascular cognitive impairment (SVCI) from patients with subcortical ischemic vascular disease (SIVD) and further stratify multidomain cognitive risk. The researchers collected DTI scans and neuropsychological scale data from an internal cohort of 134 SVCI patients and 171 SIVD patients without cognitive impairment, and used an external community cohort of 90 SVCI patients and 103 SIVD patients for unsupervised domain adaptation and independent testing. After preprocessing, DTI images were converted into white matter microstructural metrics, including FA, MD, AD, and RD, which were then fed into a DenseNet model for SVCI classification. An unsupervised domain adaptation strategy was applied to reduce distribution differences between datasets and improve model generalization on external data. The researchers then used salient maps to identify key white matter regions contributing to model decisions and computed mutual information maps between DTI images and 6 neuropsychological scales, including MMSE, MoCA, Immediate Recall, Delayed Recall, TMT-A, and TMT-B. Finally, by measuring structural similarity between each individual's salient map and domain-specific mutual information maps, and applying unsupervised clustering, SVCI patients were stratified into low-, moderate-, and high-risk subgroups for each cognitive domain.

The results showed that the DTI-based DenseNet model could accurately distinguish SVCI from SIVD patients without cognitive impairment and maintained good generalization on external data. In the internal test set, the model achieved an accuracy of 0.902; after incorporating unsupervised domain adaptation, its accuracy reached 0.926 in the target-domain test set, with an AUC of 0.942, indicating stable performance across different imaging sources. The model-generated SVCI probabilities were significantly associated with multiple neuropsychological scales, including MoCA, MMSE, Immediate Recall, Delayed Recall, and TMT-A/TMT-B, suggesting that the prediction output was not only useful for classification but also reflected cognitive impairment severity. Salient map analysis further showed that model decisions mainly relied on white matter tracts such as the corona radiata, corpus callosum, posterior limb of the internal capsule, superior longitudinal fasciculus, posterior thalamic radiation, and external capsule, with the corona radiata contributing most prominently. These regions are closely related to memory, executive function, attention, and visuospatial deficits commonly observed in SVCI. For cognitive profiling, the researchers found distinct white matter relevance patterns across different neuropsychological scales and used structural similarity between individual salient maps and mutual information maps to stratify SVCI patients into low-, moderate-, and high-risk subgroups for each cognitive domain. Patients with higher similarity showed worse cognitive performance, indicating that this framework can further support individualized multidomain cognitive risk stratification.

The significance of this work lies not only in accurately distinguishing SVCI from SIVD using DTI and DenseNet, but also in moving imaging AI beyond disease classification toward individualized cognitive risk profiling. Through unsupervised domain adaptation, the model maintained good generalization on external data; by combining salient maps with mutual information maps, the study further showed that the white matter regions highlighted by the model were neuropsychologically meaningful and could reflect impairment risks across cognitive domains such as memory, executive function, and attention. This approach provides clinicians with a more objective and scalable complementary tool, especially in settings where comprehensive neuropsychological testing is difficult to perform, and offers a new direction for precise stratification and personalized intervention in SVCI patients. At the same time, several limitations remain: the sample size is still relatively modest for deep learning training, generalizability across centers, scanners, and populations requires larger-scale validation, and the current analysis is mainly cross-sectional, meaning it cannot yet directly predict future cognitive decline at the individual level. In addition, the cognitive risk subgroups need further validation with long-term follow-up and multimodal imaging.

"Future studies incorporating larger multicenter longitudinal datasets, functional imaging, and blood biomarkers may further enhance the clinical value of this framework for precision diagnosis and intervention guidance in SVCI." said Miao He.

Authors of the paper include Miao He, Yunsi Yin, Junda Qu, Yan Wang, Xinwei Que, Xinyi Xia, Tongtong Zhang, Jiangting Li, Junyi Shen, Weihong Song, Qi Qin, Chunlin Li, and Yi Tang.