Diabetic kidney disease (DKD) affects 40% of diabetic patients, but current diagnosis relies on late-stage indicators (such as proteinuria and eGFR decline), leading to irreversible kidney damage by the time of diagnosis. The dynamic integration of multi-omics and renal imaging can reveal the early molecular-structural interaction mechanisms of DKD, breaking through the lag limitations of traditional biomarkers. By integrating the large-scale renal MRI imaging and multi-omics data of different stages of diabetic patients in the UKB, it is possible to identify the key molecular pathways driving the development of DKD; construct a deep learning model to predict high-risk groups with a >40% decline in glomerular filtration rate within 5 years by combining the renal imaging features and multi-omics data of diabetic patients at different stages, and discover biomarkers that appear earlier than proteinuria, such as combinations of plasma lipid metabolites and urine miRNA markers.
Expected outcomes include: (1) Establishing the first multi-modal data-based DKD progression risk stratification system to improve the diagnostic sensitivity for diabetic patients at different stages; (2) Developing open-source AI tools to automatically generate individualized renal injury trajectory reports based on renal imaging omics features; (3) Promoting the clinical validation and certification of new biomarkers.
