Individuals in cluster 1, with the highest vascular risk factor burden, showed significant global gray matter loss and WMH accumulation compared to clusters 2 and 3.
Advanced multi-modal deep learning identified three distinct patterns of brain volume changes associated with age and vascular risk factors, highlighting the heterogeneous impact of vascular risk on brain structure.
Absolute Event Rate: 0% vs 0%
Background: Age and vascular risk factors (VRFs) contribute to changes in brain structure which include accumulation of white matter hyperintensities (WMH), gray matter (GM) atrophy. However, data from simultaneous examination of GM and WMH volumes in relation to VRFs and aging are sparce. Therefore, we used an advanced multi-modal deep learning approach to identify synergistic brain structural changes in relation to VRFs. Methods: We used clinical brain MRIs acquired in 5,732 community-dwelling adults (age 50–76; n=3,088 women). Anyone with known neurological diagnosis or evidence of structural lesions resulting from a neurological diagnosis, was excluded. Deep Non-Negative Matrix Factorization Neural Network was jointly applied to GM volumes and total WMH burden, followed by unsupervised consensus clustering. Cluster assignments were linked to demographics and vascular risk factors (VRFs: hypertension, diabetes, dyslipidemia, coronary artery disease, smoking, BMI) by ANOVA and Chi-square testing. Results: Our analysis revealed three robust imaging subtypes (clusters). The overall difference between the clusters is shown as a heatmap which highlights GM regions and WMH volume that differentiated the clusters (Figure 1). Cluster 1 had the highest WMH burden and pronounced GM loss across multiple regions. Clusters 2 and 3 had similar WMH burden (less than cluster 1), but they differed in GM structures. Frontal and subcortical GM volumes were relatively preserved in cluster 2, while cingulate, parahippocampal, and precuneus volumes were relatively preserved in cluster 3 (Figure 2). Table 1 summarizes the VRF profile of clusters. Cluster 1 individuals were older and had a higher VRF burden than the other two clusters. Cluster 2 and 3 had similar VRF profile. Conclusion: Our findings highlight distinct patterns of GM volume loss and WMH accumulation associated with age and VRFs. While the cluster with greatest VRF burden exhibited global GM loss and highest WMH burden, other two, which had similar WMH and VRF profiles had distinct patterns of GM atrophy. Whether these GM atrophy profiles translate to distinct cognitive profiles is unknown, but the expectation is that cluster 2 would exhibit preserved executive function while and cluster 3 would exhibit better memory function. Ongoing work is targeting analyses to further phenotype these clusters and advance our understanding of underlying mechanisms selecting different patterns of GM atrophy in relation to VRFs.
Yaghoobi et al. (Thu,) reported a other. Individuals in cluster 1, with the highest vascular risk factor burden, showed significant global gray matter loss and WMH accumulation compared to clusters 2 and 3.