Kidney segmentation in neck-to-knee body MRI of 40,000 UK Biobank participants

Abstract

The UK Biobank is collecting extensive data on health-related characteristics of over half a million volunteers. The biological samples of blood and urine can provide valuable insight on kidney function, with important links to cardiovascular and metabolic health. Further information on kidney anatomy could be obtained by medical imaging. In contrast to the brain, heart, liver, and pancreas, no dedicated Magnetic Resonance Imaging (MRI) is planned for the kidneys. An image-based assessment is nonetheless feasible in the neck-to-knee body MRI intended for abdominal body composition analysis, which also covers the kidneys. In this work, a pipeline for automated segmentation of parenchymal kidney volume in UK Biobank neck-to-knee body MRI is proposed. The underlying neural network reaches a relative error of 3.8%, with Dice score 0.956 in validation on 64 subjects, close to the 2.6% and Dice score 0.962 for repeated segmentation by one human operator. The released MRI of about 40,000 subjects can be processed within one day, yielding volume measurements of left and right kidney. Algorithmic quality ratings enabled the exclusion of outliers and potential failure cases. The resulting measurements can be studied and shared for large-scale investigation of associations and longitudinal changes in parenchymal kidney volume.

MIMIR: Deep Regression for Automated Analysis of UK Biobank Body MRI

“…For some, alternative techniques may be superior, such as liver fat measurements from the dedicated UKB liver MRI with more accurate fat fraction values. Here, the projected, two-dimensional input format also limits the accuracy of organ measurements, with the proposed kidney volume measurements almost doubling the error previously achieved by segmentation of axial slices [8]. Finally, MRI-based diagnosis of type 2 diabetic status may not be clinically viable, with a specificity of 0.965 and sensitivity of only 0.250 here.…”

“…They were grouped into four modules, for each of which one network instance with one or more outputs was trained: Body composition, abdominal organs, anthropometric/experimental estimates, and age. The available reference values for these regression targets originate from previously shared atlas-based segmentation results [2], manual analyses [3], DXA imaging [1], and prior neural network segmentations of the liver and kidneys [8]. Across all subjects, 73% of these values were not available through UKB.…”

“…Information on body composition [2] and liver fat content [3] can be extracted from these images, with important implications for metabolic and cardiovascular disease [4]. Although the required image analysis can involve human labor, fully-automated approaches more recently introduced neural networks for segmentation of the liver [5], pancreas [6], [7], kidneys [8], and various other muscles, organs, and tissues within this study [9], [10]. Many of these techniques target the UKB neck-to-knee body MRI, which can represent almost all of human anatomy in one comprehensive image.…”

Reliable kidney size determination by magnetic resonance imaging in pathophysiological settings

“…Recent studies on machine learning‐based renal segmentation using neural networks reported processing times as good as 1 to 10 s per subject 76‐78 . Although these processing times are superior to our analytic approach, the effort needed to setup meticulously annotated imaging data that can be used to train, validate and test artificial intelligence algorithms must also be included in order to make a fair benchmarking of processing times.…”

Kidney Segmentation in Renal Magnetic Resonance Imaging - Current Status and Prospects

“…In the US, the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) has compiled a large database of T1w and T2w MRI images of patients with ADPKD [96], [97] fueled from the CRISP consortium and its related studies [25], [98]. Besides dedicated renal imaging databases, cohort studies like the UK Biobank or the German National Cohort might be valuable resources to further foster renal image segmentation and its evaluation [99]. However, these resources are not Open Access, e.g.…”