Improvement of MRI‐functional measurement with automatic movement correction in native and transplanted kidneys

Abstract

Purpose: To improve 2D software for motion correction of renal dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) and to evaluate its effect using the PatlakRutland model. Materials and Methods:A subpixel-accurate method to correct for kidney motion during DCE-MRI was evaluated on native and transplanted kidneys using data from two different institutions with different magnets and protocols. The Patlak-Rutland model was used to calculate glomerular filtration rate (GFR) on a voxel-by-voxel basis providing mean (K p ) and uncertainty ( ͑K p ͒) values for GFR. Results:In transplanted kidneys, average absolute variation of K p was 6.4% Ϯ 4.8% (max ϭ 16.6%). In native kidneys average absolute variation of K p was 12.11% Ϯ 6.88% (max ϭ 25.6%) for the right and 11.6% Ϯ 6% (max ϭ 20.8%) for the left. Movement correction showed an average reduction of ͑K p ͒ of 6.9% Ϯ 6.6% (max ϭ 21.4%) in transplanted kidneys, 30.9% Ϯ 17.6% (max ϭ 60.8%) for the right native kidney, and 31.8% Ϯ 14% (max ϭ 55.3%) for the left kidney. Conclusion:The movement correction algorithm showed improved uncertainty on GFR computation for both native and transplanted kidneys despite different spatial resolution from the different MRI systems and different levels of signal-to-noise ratios on DCE-MRI. VIRTUALLY ALL DISEASES of the kidney affect perfusion and glomerular filtration. Noninvasive and accurate measurement of both perfusion and glomerular filtration rate (GFR) could have a major impact in understanding renal physiopathology and for serial monitoring of the course of both acute and chronic kidney diseases. Dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) is advocated to evaluate these functional parameters. However, publications in the literature show poor correlation when MRI-GFR has been compared with GFR measured by reference methods (1,2); this poor correlation precludes the use of DCE-MRI for GFR estimation in daily clinical practice. The inaccuracy is multifactorial, with unsolved problems regarding the ideal acquisition sequence; the dual MR effect (T1 and T2*) of contrast agents; the conversion of signal intensity into concentration; the pharmacokinetic models applied; as well as the difficulties in postprocessing (segmentation and region of interest [ROI]). DCE-MRI images are usually acquired during spontaneous breathing that will result in kidney movement. Since respiratory-gated sequences (3) would lead to loss of temporal resolution, most groups studying DCE-MRI GFR have either repositioned images manually or ignored movement. However, movement causes artifacts in the pixel-based time-intensity analysis that will lead to inaccurate GFR quantification. A compromise in choosing the acquisition parameters is required to achieve a sufficiently high signal-tonoise ratio (SNR). An ideal rapid isotropic 3D imaging of the moving kidneys is not achievable, and thus slices are usually oriented along the long axis of the kidney that will allow the best approximation of movement to be estimated. During DCE-MRI, ...

Movement correction in DCE-MRI through windowed and reconstruction dynamic mode decomposition

“…From the clinical point of view, this can only be answered with a conjecture as we have no patient data but we would argue 'yes' [29]. This question provides a good opportunity to recognise that both volunteers and patients will be susceptible to motion.…”

“…Senneville et al's [29] work assumes that the kidney is a rigid body, and its shape does not change during the MRI data acquisition. For registration purposes in Senneville's work, a human expert selects a reference image from the image sequence.…”

Estimating nonrigid motion from inconsistent intensity with robust shape features

“…10 To assess renal function using DCE-MRI, pharmacokinetic models are fitted to the voxelwise enhancement curves to map the renal blood flow and obtain gleomerular filtration rate. 11 Kidney motion caused by breathing 12 induces artifacts and inaccuracy in the voxel-based analysis [13][14][15][16] and need to be corrected via image registration. On the other hand, the kinetics of the contrast agent leads to temporally inconsistent intensity, which makes estimating and subsequently correcting for the physiological and physical motion challenging.…”

A non-invasive diagnostic system for early assessment of acute renal transplant rejection.

“…These shortcomings have been circumvented recently by evaluating kidney functions with magnetic resonance imaging (MRI). For example, the dynamic contrast-enhanced (DCE) MRI has been exploited for renal function assessment due to providing both anatomical and functional kidney information [91,92,96,109]. However, because the contrast agents may cause the development of nephrogenic systemic fibrosis, many medical centers are reluctant in applying the DCE-MRI to patients with renal disease [101].…”

“…However, the kidney was segmented manually, and the evaluation of perfusion parameters (plasma volume and tubular flow) was performed visually by trained physicians for 10 data sets of healthy volunteers. Semi-automated evaluation of renal function for both native and transplanted kidneys was explored by De Senneville [92] using rigid-body registration to handle kidney motion inside a user-defined ROI. The renal cortex was segmented manually, and the GFR was estimated with Patlak-Rutland tracer kinetic model.…”