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

• DOI: 10.48550/arxiv.2006.06996
• Publication Date: 2020-01-01
• Authors: Taro Langner; Andreas Östling; Lukas Maldonis; Albin Karlsson; Daniel Olmo; D. Lindgren; Andreas Wallin; Lowe Lundin; Robin Strand; Håkan Åhlström; Joel Kullberg
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Kidney; Quantitative Biology - Quantitative Methods; Article; Machine Learning (cs.LG); 03 medical and health sciences; 0302 clinical medicine; Image Interpretation, Computer-Assisted; FOS: Electrical engineering, electronic engineering, information engineering; Humans; Knee; Quantitative Methods (q-bio.QM); Biological Specimen Banks; Image and Video Processing (eess.IV); Reproducibility of Results; Electrical Engineering and Systems Science - Image and Video Processing; Magnetic Resonance Imaging; United Kingdom; 3. Good health; FOS: Biological sciences; Radiologi och bildbehandling; Neural Networks, Computer; Algorithms; Neck; Radiology, Nuclear Medicine and Medical Imaging

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