Singular Value Decomposition Deconvolution Research Articles

Correlation between the CT Perfusion Parameter Values and Response to Recanalization in Patients with Acute Ischemic Stroke.

CT perfusion (CTP) provides various hemodynamic parameters. However, it is unclear which CTP parameters are useful in predicting clinical outcome in patients with acute ischemic stroke (AIS). Between February 2019 and June 2021, patients with anterior circulation large-vessel occlusion who achieved successful recanalization within 8 hours after stroke onset were included. The relative CTP parameter values analyzed by the reformulated singular value decomposition (SVD) method in the affected middle cerebral artery territories compared to those in the unaffected side were calculated. In addition, the ischemic core volume (ICV) was evaluated using a Bayesian Vitrea. The final infarct volume (FIV) was assessed by 24-hour MRI. The correlation between these CTP-derived values and clinical outcome was assessed. Forty-two patients were analyzed. Among the CTP-related parameters, the ICV, relative cerebral blood volume (rCBV), and relative mean transit time (rMTT) showed a strong correlation with the FIV (ρ = 0.74, p <0.0001; ρ = -0.67, p <0.0001; and ρ = -0.66, p <0.0001, respectively). In multivariate analysis, rCBV, rMTT, and ICV were significantly associated with good functional outcome, which was defined as a modified Rankin Scale score ≤2 (OR, 6.87 [95% CI, 1.20-39.30], p = 0.0303; OR, 11.27 [95% CI, 0.97-130.94], p = 0.0269; and OR, 36.22 [95% CI, 2.78-471.18], p = 0.0061, respectively). Among the CTP parameters analyzed by the SVD deconvolution algorithms, rCBV and rMTT could be useful imaging predictors of response to recanalization in patients with AIS, and the performances of these variables were similar to that of the ICV calculated by the Bayesian Vitrea.

A Machine Learning Approach to Perfusion Imaging With Dynamic Susceptibility Contrast MR.

Background: Dynamic susceptibility contrast (DSC) MR perfusion is a frequently-used technique for neurovascular imaging. The progress of a bolus of contrast agent through the tissue of the brain is imaged via a series of T2*-weighted MRI scans. Clinically relevant parameters such as blood flow and Tmax can be calculated by deconvolving the contrast-time curves with the bolus shape (arterial input function). In acute stroke, for instance, these parameters may help distinguish between the likely salvageable tissue and irreversibly damaged infarct core. Deconvolution typically relies on singular value decomposition (SVD): however, studies have shown that these algorithms are very sensitive to noise and artifacts present in the image and therefore may introduce distortions that influence the estimated output parameters.Methods: In this work, we present a machine learning approach to the estimation of perfusion parameters in DSC-MRI. Various machine learning models using as input the raw MR source data were trained to reproduce the output of an FDA approved commercial implementation of the SVD deconvolution algorithm. Experiments were conducted to determine the effect of training set size, optimal patch size, and the effect of using different machine-learning models for regression.Results: Model performance increased with training set size, but after 5,000 samples (voxels) this effect was minimal. Models inferring perfusion maps from a 5 by 5 voxel patch outperformed models able to use the information in a single voxel, but larger patches led to worse performance. Random Forest models produced had the lowest root mean squared error, with neural networks performing second best: however, a phantom study revealed that the random forest was highly susceptible to noise levels, while the neural network was more robust.Conclusion: The machine learning-based approach produces estimates of the perfusion parameters invariant to the noise and artifacts that commonly occur as part of MR acquisition. As a result, better robustness to noise is obtained, when evaluated against the FDA approved software on acute stroke patients and simulated phantom data.

Acute stroke: a comparison of different CT perfusion algorithms and validation of ischaemic lesions by follow-up imaging

To compare ischaemic lesions predicted by different CT perfusion (CTP) post-processing techniques and validate CTP lesions compared with final lesion size in stroke patients. Fifty patients underwent CT, CTP and CT angiography. Quantitative values and colour maps were calculated using least mean square deconvolution (LMSD), maximum slope (MS) and conventional singular value decomposition deconvolution (SVDD) algorithms. Quantitative results, core/penumbra lesion sizes and Alberta Stroke Programme Early CT Score (ASPECTS) were compared among the algorithms; lesion sizes and ASPECTS were compared with final lesions on follow-up MRI + MRA or CT + CTA as a reference standard, accounting for recanalisation status. Differences in quantitative values and lesion sizes were statistically significant, but therapeutic decisions based on ASPECTS and core/penumbra ratios would have been the same in all cases. CTP lesion sizes were highly predictive of final infarct size: Coefficients of determination (R (2)) for CTP versus follow-up lesion sizes in the recanalisation group were 0.87, 0.82 and 0.61 (P < 0.001) for LMSD, MS and SVDD, respectively, and 0.88, 0.87 and 0.76 (P < 0.001), respectively, in the non-recanalisation group. Lesions on CT perfusion are highly predictive of final infarct. Different CTP post-processing algorithms usually lead to the same clinical decision, but for assessing lesion size, LMSD and MS appear superior to SVDD. Following an acute stroke, CT perfusion imaging can help predict lesion evolution. Delay-insensitive deconvolution and maximum slope approach are superior to delay-sensitive deconvolution regarding accuracy. Different CT perfusion post-processing algorithms usually lead to the same clinical decision. CT perfusion offers new insights into the evolution of stroke.

Abstract 3200: Comparison of CT Perfusion Mismatch with Perfusion-Diffusion MRI in Ischemic Stroke - a Reliable Alternative?

Background and purpose: CT-perfusion (CTP) is widely and rapidly accessible for imaging acute ischemic stroke. However, there has been limited validation of CTP parameters against the more intensively studied MRI perfusion-diffusion mismatch paradigm. We tested the correspondence of CTP with contemporaneous perfusion-diffusion MRI. Methods: Acute ischemic stroke patients &lt;6hr after onset had CTP and perfusion-diffusion MRI within 1hr, before reperfusion therapies. Relative cerebral blood flow (relCBF) and time-to-peak of the deconvolved tissue-residue-function (Tmax) were calculated (standard singular value decomposition deconvolution). The diffusion lesion was registered to the CTP slabs and manually outlined to its maximal visual extent. CT-infarct core was defined as relCBF&lt;31% contralateral mean as previously published using this software. The volumetric accuracy of relCBF core compared to the diffusion lesion was tested in isolation, but also when restricted to pixels with relative time-to-peak (TTP) &gt;4sec, to reduce artifactual false positive low CBF (eg in leukoaraiosis). The MR Tmax&gt;6sec perfusion lesion (previously validated to define penumbral tissue at risk of infarction) was automatically segmented and registered to the CTP slabs. Receiver operating characteristic (ROC) analysis determined the optimal CT-Tmax threshold to match MR-Tmax&gt;6sec, confidence intervals generated by bootstrapping. Agreement of these CT parameters with MR perfusion-diffusion mismatch on co-registered slabs was assessed (mismatch ratio &gt;1.2, absolute mismatch&gt;10mL, infarct core&lt;70mL). Results: In analysis of 98 CTP slabs (54 patients, median onset to CT 190min, median CT to MR 30min), volumetric agreement with the diffusion lesion was substantially improved by constraining relCBF&lt;31% within the automated TTP perfusion lesion ROI (median magnitude of volume difference 9.0mL vs unconstrained 13.9mL, p&lt;0.001). ROC analysis demonstrated the best CT-Tmax threshold to match MR-Tmax&gt;6sec was 6.2sec (95% confidence interval 5.6-7.3sec, ie not significantly different to 6sec), sensitivity 91%, specificity 70%, AUC 0.87. Using CT-Tmax&gt;6s “penumbra” and relCBF&lt;31% (restricted to TTP&gt;4s) “core”, volumetric agreement was sufficient for 90% concordance between CT and MRI-based mismatch status (kappa 0.80). Conclusions: Automated CTP mismatch classification using relCBF and Tmax is similar to perfusion-diffusion MRI. CTP may allow more widespread application of the “mismatch” paradigm in clinical practice and trials.

Tracer Delay Correction of Cerebral Blood Flow with Dynamic Susceptibility Contrast-Enhanced MRI

Cerebral blood flow (CBF) and vascular mean transit time (MTT) can be determined by dynamic susceptibility contrast-enhanced magnetic resonance imaging and deconvolution with an arterial input function. However, deconvolution by a singular value decomposition (SVD) method is sensitive to the tracer delay that often occurs in patients with cerebrovascular disease. We investigated the effect of tracer delay on CBF determined by SVD deconvolution. Simulation study showed that underestimation of CBF due to tracer delay was larger for shorter MTTs. We developed a delay correction method that determines tracer delay by means of least-squares fitting pixel-by-pixel. The corrected CBF was determined by SVD deconvolution after time-shifting of the measured concentration curve. The simulations showed that the corrected CBF was insensitive to tracer delay irrespective of the vascular model, although CBF fluctuation increased slightly. We applied the delay correction to the CBF and MTT images acquired for nine patients with hyperacute stroke and unilateral occlusion of the middle cerebral artery. We found in some patients that the delay correction modulated the contrast of CBF and MTT images. For hyperacute stroke patients, tracer delay correction is essential to obtain reliable perfusion image when SVD deconvolution is used.

Measurement of cerebral circulation by dynamic susceptibility contrast-enhanced MRI: effect of tracer delay

Abstract Cerebral blood flow (CBF) and mean transit time (MTT) can be determined by dynamic susceptibility contrast-enhanced magnetic resonance imaging (DSC-MRI) and deconvolution with an arterial input function (AIF). However, deconvolution with a singular value decomposition (SVD) method is sensitive to the tracer delay that often occurs in patients with cerebrovascular disease. We developed a delay correction method that determines tracer delay by means of least-squares fitting pixel-by-pixel. The corrected CBF was determined by SVD deconvolution after time-shifting of the measured concentration curve. The simulations showed that the corrected CBF was insensitive to tracer delay irrespective of the vascular model, although CBF fluctuation increased slightly. We applied the delay correction to the CBF and MTT images acquired for a patient with hyperacute stroke and unilateral occlusion of the middle cerebral artery. The delay correction modulated the contrast of perfusion images. For hyperacute stroke patients, tracer delay correction is essential to obtain reliable CBF and MTT images when SVD deconvolution is used.

Aspects on the accuracy of cerebral perfusion parameters obtained by dynamic susceptibility contrast MRI: a simulation study.

Several studies have indicated that deconvolution based on singular value decomposition (SVD) is a robust concept for retrieval of cerebral blood flow in dynamic susceptibility contrast (DSC) MRI. However, the behavior of the technique under typical experimental conditions has not been completely investigated. In the present study, cerebral perfusion was simulated using different temporal resolutions, different signal-to-noise ratios (S/Ns), different shapes of the arterial input function (AIF), different signal drops, and different cut-off levels in the SVD deconvolution. Using Zierler's area-to-height relationship in combination with the central volume theorem, calculations of regional cerebral blood volume (rCBV), regional cerebral blood flow (rCBF), and regional mean transit time (rMTT) were accomplished, based on simulated DSC-MRI signal curves corresponding to artery, gray matter (GM), white matter (WM), and ischemic tissue. Gaussian noise was added to the noise-free signal curves to generate different S/Ns. We studied image time intervals of 0.5, 1.0, 1.5, 2.0, 2.5, and 3.0 s, as well as different degrees of signal decrease. The singular-value threshold in the SVD procedure and the shape of the AIF were also varied. Increased rCBF was seen when noise was added, especially for rCBF in WM at the larger image time intervals. The rCBF showed large standard deviations using a low threshold value. A prolonged time interval led to a lower absolute value of rCBF both in GM and WM, and a low/broad AIF also underestimated the rCBF. When a larger maximal signal decrease was assumed, smaller standard deviations were observed. No systematic change of the average rCBV was observed with increasing noise or with increasing image time interval. At S/N = 40, a low cut-off value resulted in an rCBF that was closer to the true value. Furthermore, at low S/N it was difficult to differentiate ischemic tissue from WM.

Monte-Carlo error analysis in x-ray spectral deconvolution

The deconvolution of spectral information from sparse x-ray data is a widely encountered problem in data analysis. An often-neglected aspect of this problem is the propagation of random error in the deconvolution process. We have developed a Monte-Carlo approach that enables us to attach error bars to unfolded x-ray spectra. Our Monte-Carlo error analysis has been incorporated into two specific deconvolution techniques: the first is an iterative convergent weight method; the second is a singular-value-decomposition (SVD) method. These two methods were applied to an x-ray spectral deconvolution problem having m channels of observations with n points in energy space. When m is less than n, this problem has no unique solution. We discuss the systematics of nonunique solutions and energy-dependent error bars for both methods. The Monte-Carlo approach has a particular benefit in relation to the SVD method: It allows us to apply the constraint of spectral nonnegativity after the SVD deconvolution rather than before. Consequently, we can identify inconsistencies between different detector channels.