Assessment of 3D DCE-MRI of the kidneys using non-rigid image registration and segmentation of voxel time courses

https://doi.org/10.1016/j.compmedimag.2008.11.004Get rights and content

Abstract

We have applied automated image analysis methods in the assessment of human kidney perfusion based on 3D dynamic contrast-enhanced MRI data. This approach consists of non-rigid 3D image registration of the moving kidney followed by k-means clustering of the voxel time courses with split between left and right kidney. This method was applied to four data sets acquired from healthy volunteers, using 1.5 T (2 exams) and 3 T scanners (2 exams).

The proposed registration method reduced motion artifacts in the image time series and improved further analysis of the DCE-MRI data. The subsequent clustering to segment the kidney compartments was in agreement with manually delineations (similarity score of 0.96) in the same motion corrected images. The resulting mean intensity time curves clearly show the successive transition of contrast agent through kidney compartments (cortex, medulla, and pelvis). The proposed method for motion correction and kidney compartment segmentation might improve the validity and usefulness of further model-based pharmacokinetic analysis of kidney function in patients.

Introduction

The kidneys maintain normal homeostasis by filtering and excreting metabolic waste products, by regulating acid–base balance, and by moderating blood pressure and fluid volume [1]. Decrease in renal function is caused by many disorders, among these are diabetes mellitus and hypertension. Chronic renal failure is an increasing problem world-wide; up to 5% of the world’s population may in the near future suffer from end-stage renal disease (ESRD), with dialysis or kidney transplantation as the costly therapeutic alternatives [2]. Furthermore, renovascular disease seems to be an individual risk factor for cardiovascular disease [3]. Therefore, it is important – for patients and society – that methods are developed to monitor renal function precisely, thus enhancing the assessment of disease progression, the prognosis, and follow-up therapy.

At present, diagnosis of renal dysfunction is based on such measurements as creatinine, urea, and electrolytes, as well as on creatinine clearance. These indirect measurements have low sensitivity, since a significant change in creatinine level is only detectable after a 60% function loss has occurred, while creatinine clearance overestimates the actual glomerular filtration rate (GFR) by up to 20% [4]. In addition, these clinical chemistry measurements cannot detect local differences in the kidneys and cannot distinguish between left and right kidney. To overcome these limitations, dynamic contrast-enhanced MR imaging (DCE-MRI) has emerged as a technique that can be used for the more accurate assessment of regional renal function [5], [6]. With this technique, signal intensity evolution can be measured and visualized as images that reflect the passage of an injected tracer or contrast agent through the organ.

An important obstacle to these dynamic measurement techniques that complicates further analysis is the movement of the organ of interest during image acquisitions, when the individual voxels undergo complex displacements due to respiratory motion and pulsation. Such movements are often overlooked in studies of renal function [7], [8], [9]. However, without proper motion correction, the derived voxel time courses will not represent spatially fixed kidney volume elements, thus invalidating a major assumption underlying subsequent voxel-based time series analysis and pharmacokinetic modeling. A major contribution of the present work is that we performed geometric correction of these movements by introducing multi-modality non-rigid image registration techniques. The use of these techniques can improve the applicability and clinical value of recent developments in renal multi-compartment modeling, in simulation, and in image-based estimation of renal function [10], [11], [12], [13].

Another important issue in analyzing kidney function using DCE-MRI is the detection and segmentation of the renal compartments. Accurate parenchyma sub-segmentation enables the assessment of signal intensity time curves for those regions, leading to a more comprehensive evaluation of the status of the organ and of its functional compartments [14], [10].

The assessment of MRI time series is usually carried out manually or semi-automatically [6], [15], [16]. Typically, the user delineates a region of interest (ROI) from which a (mean) voxel time course is extracted. In this way the ROIs are generally selected based on the user’s knowledge of anatomy. Valuable information inherent in the signal intensity time courses is not used. Manual ROI placements are also subject to inter- and intra-observer variability [15]. An additional disadvantage of such methods is that they are slow, even though semi-automatically techniques do reduce the processing time [17]. Automated computational techniques can overcome these drawbacks. Unsupervised data-driven approaches such as clustering of the voxel time courses may lead to more accurate and objective segmentation within reasonable time.

In this paper we present an automated 3D non-rigid image registration and pattern recognition technique, applied to dynamic contrast-enhanced MR imaging, to extract useful voxel-based functional information from the kidney.

Section snippets

Image registration

During the past three decades several registration techniques have been developed and applied to medical imaging. Comprehensive surveys are presented in [18], [19]. In the present work we have focused on non-rigid (deformable) registration methods relevant in cases where the imaged object can become deformed during the observation time. A high degree of deformation and displacement is especially prominent in dynamic cardiac imaging, breast imaging, and abdominal imaging.

Motion correction of

Data and acquisition

In the present study we have used two different pulse-sequences for the acquisition of 3D perfusion time series having a different temporal and spatial resolution as well as a different length. On the 1.5 T Siemens Symphony scanner and 3.0 T Sigma Excite GE we used a 3D volumetric interpolated breath-hold examination (VIBE) and a 3D liver acceleration volumetric acquisition (LAVA), respectively. Altogether, four data sets were obtained in four exams; each exam resulted in one data set. Table 1

Results

In this section the results of the proposed algorithms obtained for the four data sets (cf. Section 3.1 and Table 1) are presented. The overall assessment of the proposed method consists of three steps: (i) assessment of the registration method, (ii) analysis of the impact of registration to clustering, and (iii) assessment of k-means based segmentation compared to manual segmentation.

Discussion

In this paper we have introduced a set of automated methods for the assessment of renal function from 3D DCE-MRI acquisitions, comprising multi-modal, non-rigid 3D image registration and k-means clustering that together enable the extraction of valid voxel time courses from recorded data. Up to now, the problem of kidney motion has generally been ignored when analyzing dynamic image data using model-free descriptive techniques or pharmacokinetic models [7], [8], [9], [10], [11]. However,

Frank G. Zöllner received the diploma and a PhD degree (Dr.-Ing.) in computer science from the University of Bielefeld, Germany, in 2001 and 2004, respectively. He joined the Applied Computer Science group in 2001 and worked towards his PhD within the bioinformatics graduate program until 2004. From 2004 until 2006 he worked as a post-doctoral researcher in the BMBF project ALPIC. From February to May 2006 he was a guest researcher at the Computational Biology Unit, Bergen Center for

References (45)

  • European Kidney Health Association. Factsheet: the kidney in health and disease. Last accessed on 9.8.2008; 2008....
  • Myers GL, Miller WG, Coresh J, Fleming J, Greenberg N, Greene T, et al. Recommendations for improving serum creatinine...
  • P.V. Prasad

    Functional MRI of the kidney: tools for translational studies of pathophysiology of renal disease

    Am J Physiol Renal Physiol

    (2006)
  • M. Dujardin et al.

    Quantification of renal perfusion and function on a voxel-by-voxel basis: a feasibility study

    Magn Reson Med

    (2005)
  • D.L. Buckley et al.

    Measurement of single kidney function using dynamic contrast-enhanced MRI: comparison of two models in human subjects

    J Magn Reson Imag

    (2006)
  • N. Michoux et al.

    Analysis of contrast-enhanced MR images to assess renal function

    Magn Reson Mater Phys

    (2006)
  • V.S. Lee et al.

    Renal function measurements from mr renography and a simplified multicompartmental model.

    Am J Physiol Renal Physiol

    (2007)
  • S.P. Sourbron et al.

    MRI-measurement of perfusion and glomerular filtration in the human kidney with a separable compartment model

    Invest Radiol

    (2008)
  • H.J. Michaely et al.

    Renal artery stenosis: functional assessment with dynamic MR perfusion measurements—feasibility study

    Radiology

    (2006)
  • J.A. de Priester et al.

    MR renography: an algorithm for calculation and correction of cortical volume averaging in medullary renographs

    J Magn Reson Imag

    (2000)
  • C.H. Coulam et al.

    Measurement of renal volumes with contrast-enhanced MRI

    J Magn Reson Imag

    (2002)
  • H. Rusinek et al.

    Performance of an automated segmentation algorithm for 3D MR renography

    Magn Reson Med

    (2007)
  • Cited by (106)

    • Quantitative imaging for radiotherapy purposes

      2020, Radiotherapy and Oncology
    • CorteXpert: A model-based method for automatic renal cortex segmentation

      2017, Medical Image Analysis
      Citation Excerpt :

      For registration-based methods, Sun et al. (2004) presented an image registration algorithm to delineate the renal cortex for dynamic renal perfusion MR images. Another registration-based approach (Zöllner et al., 2009) separated the inner compartments of kidney by k-means clustering after a non-rigid registration. Lately, Song et al. (2008) presented a 4D level set framework for dynamic MR images kidney segmentation.

    View all citing articles on Scopus

    Frank G. Zöllner received the diploma and a PhD degree (Dr.-Ing.) in computer science from the University of Bielefeld, Germany, in 2001 and 2004, respectively. He joined the Applied Computer Science group in 2001 and worked towards his PhD within the bioinformatics graduate program until 2004. From 2004 until 2006 he worked as a post-doctoral researcher in the BMBF project ALPIC. From February to May 2006 he was a guest researcher at the Computational Biology Unit, Bergen Center for Computational Sciences, University of Bergen. From May 2006 until December 2007 he was a researcher at the Section of Radiology, Institute for Surgical Sciences, Haukeland University Hospital and the Neuroinformatics and Image Analysis Group, Department of Biomedicine at the University of Bergen, Norway. Currently, he holds a 10% research position at the Section of Radiology, Institute for Surgical Sciences, Haukeland University Hospital. In 2008, Dr. Zöllner joined the Chair of Computer Assisted Clinical Medicine as a post-doctoral researcher. His research interests lie in the fields of pattern recognition, image processing as well as bioinformatics. In particular, he is interested in applying computational methods from pattern recognition, image analysis, or bioinformatics to the fields of molecular imaging and medical image analysis. Currently, he works on magnetic resonance imaging (MRI) of the human kidney for diagnostics of renal disease. Dr. Zöllner is a member of the IEEE Computer Society and the European Society for Magnetic Resonance in Medicine and Biology (ESMRMB).

    Rosario Sance received a MEng degree in 2005 from Univ Politécnica de Madrid (UPM), with a master thesis on “MRI registration for the evaluation of the renal function”. Currently she is working for a PhD thesis in the Biomedical Image Technology group (BIT), included in the Electronic Eng. Dept. (UPM). In the framework of the project COSTB21 (Physiological Modelling of MR Image Formation), she has been working in the analysis of kidney function in collaboration with University of Bergen (Norway). Her research interests are medical image analysis, mainly image registration and dynamic image analysis.

    Peter Rogelj received a PhD degree in electrical engineering from the University of Ljubljana (ULJ), Slovenia, in 2003. He joined the Machine Vision Group at ULJ in 1998. During his study towards a master’s degree (2001) and the PhD degree, he also worked as a guest researcher at Czech Technical University, Prague, CZ, and University of Pennsylvania, Philadelphia, USA. Later he conducted several research projects in ULJ and in industry. His main research interest is image processing, with a special focus on multi-modality and non-rigid medical image registration.

    María J. Ledesma-Carbayo graduated in telecommunication engineering in 1998, with a master thesis on medical image analysis. Previously, in 1997/1998, she followed the ERASMUS European Course on Biomedical Engineering and Medical Physics in Univ. Patras (Greece). She is now associate professor in the Electronics Eng. Dept. (Universidad Politécnica de Madrid), and she has obtained her PhD degree in biomedical imaging several years ago. In 1999, she has visited the Medical Vision Lab. at Oxford Univ. during three months and in 2000 and 2001, she has collaborated with École Polytechnique Fédéral in Lausanne (Switzerland). Recently she has spent six months on the Cardiac Energetics Lab. at the National Institutes of Health (USA). Her main research areas of interest are medical image analysis and processing, especially those topics dealing with registration, cardiac imaging and motion estimation and compensation. Her research in medical imaging processing has been published in journals and conferences (more than 50). She has a lot of experience participating in research projects, from the Spanish Finantial Offices, International Agencies and International Companies. Member of IEEE.

    Jarle Rørvik has a position as associate professor in radiology at the University of Bergen and physician at the Department of Radiology at Haukeland University Hospital, Bergen, Norway.

    In 1998 he finished his thesis on use of imaging for staging organ-confined prostate cancer prior to radical treatment. Later he has continued his research on prostate cancer both for detection and staging. Currently, he and his research team perform studies on use of MRI with contrast and MR spectroscopy for both detection and staging of prostate cancer.

    Further, he has together with Arvid Lundervold established a research-group working with MRI for kidney imaging including both morphologic and functional exams. He has been supervisor for two PhD-students (doctores medicalis) and several masterstudents with a wide spectrum of research topics.

    He is in charge of the education of medical students in radiology and has received several prizes for the quality in the teaching. A special interest has been developement and use of digital teaching material for the internet.

    Andrés Santos has a PhD and a MEng from Universidad Politécnica Madrid (UPM). He has a position as professor in the Electronic Engineering Dept. (UPM), where he is the Director of the Biomedical Image Technology group since it was founded in 2000. During 1987–1988 he was International Research Fellow at SRI International (Menlo Park, California). He has been co-director of a master course in biomedical enginering and since 1995 he is also professor in the European Course on Biomedical Engineering (Univ. Patras, Greece).

    His research interests are related to biomedical image acquisition and analysis. Coordinator of several national and international projects on biomedical imaging and author or co-author of more than 150 international publications. Member of SPIE, IEEE and ESEM.

    Arvid Lundervold has a BSc in mathematics and got his medical training (MD) from University of Oslo. He has a PhD on “Multispectral analysis, classification and quantification in medical magnetic resonance imaging” from University of Bergen in 1995. He has been a research scientist at the Norwegian Computing Center in Oslo, working on image analysis and pattern recognition, and later associate professor at Department of Physiology, University of Bergen. From September 2005 he has been professor in medical information technology at Department of Biomedicine, University of Bergen and is head of the Neuroinformatics and Image Analysis Laboratory. He is also member of the Molecular Imaging Center, UoB, the Bergen fMRI group, and the interdisciplinary Bergen Image Processing group, located at the Department of Mathematics. He his participating in several national and international projects, and member of IEEE Computer Society, ISMRM, MAA and the Norwegian Medical Association.

    View full text