Epidemiologic, genetic and translational analyses of radiomics-based kidney features of two large, population-based studies

• Applicants: Dr. Elias Kellner; Professorin Dr. Anna Köttgen; Dr. Maximilian Russe
• Subject Area: Nuclear Medicine, Radiotherapy, Radiobiology ; Epidemiology and Medical Biometry/Statistics ; Medical Physics, Biomedical Technology ; Nephrology
• Term: since 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 428212052

Kidney function and damage is estimated in clinical care based on markers from routine biochemistry tests, serum creatinine as well as urine albumin and creatinine. These traditional markers are used to diagnose and stage chronic kidney disease (CKD), which affects 10% of the adult population. However, traditional markers of kidney function have limitations, including the influence of non-kidney related factors as well as limited risk prediction. Imaging-based markers of kidney function and disease are a promising, complementary source of information In the first funding period of the SPP 2177, we have therefore used abdominal MR images from >11,000 participants of the large, population-based German National Cohort (NAKO) study to train and apply a convolutional neural network for the (semi-)automated prediction of imaging markers of the kidneys (e.g., total, cortex, medulla and hilus volumes; cysts), studied their distribution, and correlated them to traditional kidney function markers and diseases. Using multi-variable adjusted regression analyses, we were able to confirm known correlates such as between lower estimated glomerular filtration rate and lower total kidney volume, as well as to identify numerous interesting new relationships, such as a correlation between the presence of diabetes and higher cortex volumes (consistent with glomerular hyperfiltration) or a differential association between medullary and hilus volumes with higher age. Based on this successful preliminary work, we propose to address three scientific aims in a second funding period of the SPP 2177: i.) Extension of the developed workflow to the full set of NAKO participants with MR images (N>30,000), identification of clinical correlates of imaging markers of the kidney, and application of the workflow to MR images from an independent, second populationbased study, the UK Biobank (N=50,000), in which we will aim to validate associations with clinical correlates. ii.) Conduct of genomewide association studies of imaging markers of the kidney, including kidney compartments and more advanced radiomics features in the UK Biobank. iii.) Forward (clinical implementation, decision support software) and reverse (transfer learning for improved segmentation) translational studies, based on both, data from the population-based NAKO obtained without contrast and on clinical images with contrast obtained during patient care. Together, the proposed aims will generate new, complementary markers of kidney function and disease that will aid the diagnosis, classification and monitoring of CKD, a public health burden of rising incidence. The developed workflows will promote interactions and provide training opportunities within the Radiomics community both nationally and internationally.

• DFG Programme: Priority Programmes
• Subproject of: SPP 2177: Radiomics: Next Generation of Biomedical Imaging
• Co-Investigators: Dr. Martin Büchert; Dr. Wilfried Reichardt; Dr. Marco Reisert; Dr. Peggy Sekula