An alternative viewpoint of the similarities and differences of SVD and FT deconvolution algorithms used for quantitative MR perfusion studies

Self Cite

Purpose:To demonstrate the degree of the cerebral blood flow (CBF) estimation bias that could arise from distortion of the arterial input function (AIF) as a result of partialvolume effects (PVEs) in dynamic susceptibility contrast (DSC) magnetic resonance imaging (MRI). Materials and Methods:A model of the volume fraction an artery occupies in a voxel was devised, and a mathematical relationship between the amount of PVE and the measured baseline MR signal intensity was derived. Based on this model, simulation studies were performed to assess the impact of PVE on CBF. Furthermore, the effectiveness of linear PVE compensation approaches on the concentration function was investigated. Results:Simulation results showed a nonlinear relationship between PVE and the resulting CBF measurement error. In addition to AIF underestimation, PVE also causes distortions of AIF frequency characteristics, leading to CBF errors varying with mean transit time (MTT). An uncorrected AIF measured at a voxel with a partial-volume fraction of Յ50% could produce a CBF overestimation of more than fourfold. Linear compensation of the concentration curves did not produce correct CBF estimates.Conclusion: PVE can induce significant CBF estimation biases. In addition, the MTT dependence of CBF accuracy raises doubts of the validity of adopting a single crosscalibration factor (i.e., setting normal white matter to 22 mL minute -1 (100 g) -1 ) to obtain CBF values with absolute units. The impact of PVE may be reduced by decreasing the maximum arterial signal drop in the perfusion images. To correct the AIF distortions introduced by PVE, the nonlinear relationship between the impact of PVE on MR signal intensity and contrast concentration function must be considered.

Section: Methodsmentioning

confidence: 99%

The impact of partial‐volume effects in dynamic susceptibility contrast magnetic resonance perfusion imaging

Chen

Smith

Frayne

2005

Self Cite

“…SDNR=30), there was good agreement between the actual and estimated flow values for MMSE ( Figure 4a) and SVD (Figure 4c) method, except for higher flow values SVD appears to be underestimating flow and FT based MMSE method appears to be more stable than SVD. The instability in SVD procedure may arise from the fact that the behavior of SVD algorithm changes for each deconvolution performed, resulting in being equivalent to FT based MMSE at times, while sometimes generating high frequency noise leakage and low frequency signal cut-off [15]. This behavior of SVD technique has been attributed to the fact that the SVD algorithm does not behave as a low pass filter but as a series of band pass filters that depend on the characteristics of noisy arterial concentration curve used during deconvolution process.…”

Section: Discussionmentioning

confidence: 99%

Cerebral blood flow estimation from perfusion-weighted MRI using FT-based MMSE filtering method

Sakoğlu

Sood

2008

Introduction-Perfusion weighted MRI can be used for estimating blood flow parameters using bolus tracking technique based on DSC MRI. In order to extract flow parameters, several deconvolution techniques have been proposed, of which, the SVD and FT based techniques are more popular and widely used. In this work, an FT based method has been proposed that involves derivation of an optimal shaped filter (defined as a filter function) estimated using minimum mean-squared error (MMSE) technique in the frequency domain. The proposed technique has been compared with the well established SVD technique using simulation experiments.

“…First, LRFs reduce the effect of the ambient noise level on deconvolution by averaging many signals from large WM ROIs and they are less susceptible to temporal aliasing, which relaxes the need for filtering to ensure deconvolution stability. This has the benefit of improving the accuracy of LRF CBF estimates by reducing MTT-dependent CBF errors resulting from deconvolution noise filtering (22,29). Second, more reliable perfusion estimates could be obtained by optimizing existing perfusion protocols in order to improve the measurement of tissue signals without the need to balance the conflicting constraints imposed by arterial signals (e.g., avoiding arterial saturation) (19).…”

Section: Discussionmentioning

confidence: 99%

“…However, these advanced methods have not yet been widely adopted in clinical practice, in part because they are new, but also because they complicate the perfusion methodology in clinical practice and, when used individually, only address AIF PVE error, obviating the need for the extra effort. For example, AIF problems such as MTT-dependent p svd CBF underestimation (22,29), and the fact that T 2 *-weighted arterial signal concentration estimation is unpredictable (except when the artery is oriented parallel to the main magnetic field, which is not readily testable in vivo) (20), are not corrected by these techniques. Additionally, correcting AIF PVE by scaling the AIF to the signal from a vein without PVE (i.e., a venous output function selected from the superior sagittal sinus) (31) does not seem feasible with DSC-MR because of signal saturation and large susceptibility artifacts that occur in large, extracranial vessels (e.g., internal carotid arteries)-especially at 3 T, which has been our experience.…”

Section: Discussionmentioning

confidence: 99%

Cerebral blood flow estimation in vivo using local tissue reference functions

Kosior¹,

Smith²,

Kosior³

et al. 2008

Self Cite

Purpose:To evaluate the use of bolus signals obtained from tissue as reference functions (or local reference functions [LRFs]) rather than arterial input functions (AIFs) when deriving cross-calibrated cerebral blood flow (CBF CC ) estimates via deconvolution. Materials and Methods:AIF and white matter (WM) LRF CBF CC maps (cross-calibrated so that normal WM was 23.7 mL/minute/100 g) derived using singular value decomposition (SVD) were examined in 28 ischemic stroke patients. Median CBF CC estimates from normal gray matter (GM) and ischemic tissue were obtained.Results: AIF and LRF median CBF CC estimates resembled one another for all 28 patients (average paired CBF CC difference 0.4 Ϯ 1.7 mL/minute/100 g and -0.4 Ϯ 1.4 mL/ minute/100 g in GM and ischemic tissue, respectively). Wilcoxon signed-rank comparisons of patient median CBF CC measurements revealed no statistically significant differences between using AIFs and LRFs (P Ͼ 0.05). Conclusion:If CBF is quantified using a patient-specific cross-calibration factor, then LRF CBF estimates are at least as accurate as those from AIFs. Therefore, until AIF quantification is achievable in vivo, perfusion protocols tailored for LRFs would simplify the methodology and provide more reliable perfusion information.