Approved research

Uncovering the heterogeneity and spatial complexity of cerebral small vessel disease with deep learning method based imaging and genetic signatures

Lay summary

Cerebral small vessel disease (CSVD) is a highly heterogeneous disease, modified by underlying genetic determinants, as well as risk, environmental and lifestyle factors. CSVD display multiple distinct brain phenotypes and spatial distribution across individuals with different risk factors and susceptibility genes, reflecting subtypes of CSVD. However, current treatments for CSVD focus mainly on antiplatelet therapy which overlooks the heterogeneity in imaging phenotypes and genetic predispositions, lacking personalized treatment strategies. Recently, the combination of artificial intelligence (AI) methods with imaging and genetic data has allowed us to dissects CSVD heterogeneity, thereby conferring genetic correlations to the CSVD subtypes and associated endophenotypic signatures.

Therefore, in this study, I intend to utilize deep learning method combined genetic and imaging data to classify patients of CSVD with different etiologies. I will focus on the aims:

1) identify subtypes of CSVD with deep learning model based imaging and genetic signatures;

2) discover genetic traits, biomarkers and cognitive features associated with these subtypes;

3) classify CSVD subjects with different risk factors and extract the genetic, endophenotypic signatures correlations to specific etiologies of CSVD;

4) develop classification tools to screen for CSVD with different risk factors;

5) study the interaction of CSVD with other organs under diseases state;

6) finally, explore the potential strategies for personalized treatment by screening medicine history and lifestyles related to prognosis of CSVD with specific etiology.

By leveraging deep learning methods alongside multimodal brain imaging and genomics, this study aims to uncover the heterogeneity and spatial complexity of CSVD, which has been largely treated as a homogeneous entity in current clinical practices. The identification of specific imaging phenotypes, genetic characteristics for subtypes of CSVD will help the development of personalized treatment plans. Furthermore, by mapping the spatial distribution of imaging features, identifying genetic traits and biomarkers, exploring interaction with other organs under diseases state, this research could provide insights into the pathophysiology and progression of CSVD.

This study will be conducted for 36 months with an excellent server. I hope that our proposed system could facilitate the understanding of heterogeneity and spatial complexity of CSVD and clinical decision.

The codes for image processing, model architecture will be open-sourced with my paper and thesis. Also, I hope the automatic and accurate segmentation models and results of brain imaging markers of CSVD could be used in other research projects, and I would be pleasure to upload this part of data to UK Biobank.