Kidney segmentation for quantitative analysis applying MaskRCNN architecture

Mathilde Overgaard Lauersen, Büsra Köylü, Bryan Haddock, John Aa. Sørensen

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

Results of applying the deep learning Mask Regional Convolutional Neural Network (MaskRCNN) to the problem of segmenting human kidney and delineating the kidney tissue into the 12 classes left/right kidney, cortex, medulla 1, 2, 3, 4 are presented, using a 3T MRI scanner. This problem originates from advances in magnetic resonance imaging MRI techniques to monitor blood perfusion and blood oxygenation in kidneys of persons suffering from reduced kidney function. This segmentation allows the processing of the numerous MRI arterial spin labelling (ASL) images and T2* images with respiration movement to calculate and map renal perfusion and evaluate oxygenation respectively. The quality of the segmentation is assessed visually and by the segment label classifier. In total, a selection of 37 segmentation experiments, combining MRI sequences and kidney tissue types are presented. To the best of the author’s knowledge, applying this segmentation method on kidney tissue is new.
Original languageEnglish
Title of host publicationProceedings of the IEEE Symposium Series on Computational Intelligence 2021
Number of pages6
PublisherIEEE
Publication statusAccepted/In press - 2022
EventIEEE Symposium Series on Computational Intelligence 2021 - Online event
Duration: 5 Dec 20217 Dec 2021

Conference

ConferenceIEEE Symposium Series on Computational Intelligence 2021
CityOnline event
Period05/12/202107/12/2021

Keywords

  • MaskRCNN
  • Segmentation
  • Human kidney tissue
  • Renal cortex
  • Medullary
  • Development path

Fingerprint

Dive into the research topics of 'Kidney segmentation for quantitative analysis applying MaskRCNN architecture'. Together they form a unique fingerprint.

Cite this