Perfusion Settings (options)

More options relating to the perfusion analysis can be accessed from the <Options> button at the bottom of the window.

The following menu will appear:

Analysis

  • Normalization: this will normalize the CBV and CBF maps to globally determined mean value. The global mean value is calculated from all pixels determined to represent 'normal' brain tissue (excluding noisy and otherwise 'abnormal' dynamic curves). See Emblem and Bjornerud AJNR Am J Neuroradiol. 2009 Nov;30(10):1929-32. doi: 10.3174/ajnr.A1680 for more details on the method. This normal brain reference mask can be either grey matter, white matter or both, and is selected in the Segmentation tab of the Advanced Options menu. The reference mask segmentation is automatically performed by the software. You can choose to generate the segmented mask as an output map. This can be turned on in the Output tab of the Options menu. Note that the MTT values will also be normalized when this option is selected.
  • Vessel segmentation: Automatic removal of vessel from perfusion maps. The pixels classified as vessels are removed (pixel value set to zero) from the output perfusion maps. The method is based on a cluster analysis of the estimated perfusion related parameters to separate vessels (both arteries and veins) from other tissues. For details of the method used, see Emblem et al. Magn Reson Med 2009 May;61(5):1210-7). NOTE: The vessel segmentation functionality is meant as an aid in identifying vessels in perfusion maps and no claims are made as to the accuracy of the method to truly identify vessels.
  • Miscellaneous: Remove outliers (spikes) from perfusion maps: When gamma variate fitting is applied, or when the input data is generally noise, the resulting perfusion values can, under certain conditions, become artificially high if the fitting algorithm fails resulting in 'spikes' in the perfusion maps. If this option is selected nordicICE will attempt to detect outliers in the output pixel values and remove these (i.e. by setting the outlier pixels equal to the maximum of the 'normal' pixel range).


Output

Use this menu to select which additional output maps you want to generate during the perfusion analysis.

  • Variance map: The variance of time curve after baseline.
  • Brain mask: The segmented normal brain tissue, either white matter, grey matter or both, as set in Advanced Options - Segmentation. Creates a binary image where each pixel is set to 1 or 0 depending on whether it is classified as normal tissue or not.
  • Vessel mask: Creates a binary image where each pixel is set to 1 or 0 depending on whether it is classified as vessel or not. 
  • Mean baseline image
  • Non-corrected CBV map: Option to create an additional CBV map where leakage correction is not done (requires that the leakage correction option is turned on). 
  • Non-segmented CBV map: Option to create an additional CBV map where vessel segmentation is not done (requires that the vessel removal option is turned on).
  • Chi-square map: This image represents the 'goodness of fit' of the raw data to the gamma variate model function for each pixel (requires that gamma variate fitting is done).
  • Regularization index map: This image gives the regularization parameter in the Tikhonov regularization (requires that Tikhonov regularization is used, see advanced options for details).

 

Deconvolution

Singular Value Deconvolution

The deconvolution procedure in nordicICE uses a mathematical technique called singular value deconvolution (SVD), first suggested for use in MRI perfusion analysis by Østergaard et al (Magn Reson Med 36:1996). One critical requirement in SVD based deconvolution is proper 'regularization' to obtain a stable solution and hence a robust estimation of the perfusion related parameters. The SVD method requires a threshold to be defined that specifies which of the components of the decomposition process represent noise and hence should be eliminated from the analysis. Different methods are available for determining the optimal threshold. A low cutoff makes the solution more sensitive to noise but gives more correct perfusion values in the noiseless case. A large cutoff value provides a more robust solution in the presence of noise but may also filter out relevant information from the AIF response which can result in an over-estimation of perfusion.

Two SVD methods are included:

  • Standard (sSVD).
  • Delay insensitive SVD. To be used if AIF peak is delayed in time compared to the tissue response.

Two regularization methods are included:

  • Fixed threshold: a set, fixed value is used. Components (of the diagonal matrix from the SVD)  less than the specified fraction (relative to the largest value) are set to zero. The threshold should be a fraction (of the maximum singular value) between zero and one.  The default value is 0.2. A larger value will give more severe filtering and may result in under-estimation of perfusion, whereas a too low value will introduce too much noise and resulting spikes in the perfusion maps. 
  • Iterative using Tikonov regularization. The optimal filter threshold value is determined iteratively by finding the 'optimal' trade-off between a 'correct' solution and an oscillating solution. For details of the method, see Hansen HC (SIAM Journal on Scientific Computing 1993;14(6): 1487 – 1503). The number of iterations to run can be specified in Advanced perfusion settings. More iterations increase the processing time.

 See Bjornerud and Emblem J Cereb Blood Flow Metab. 2010 May;30(5):1066-78. doi: 10.1038/jcbfm.2010.4 for details.

 

Output scaling factor

An output scaling factor is used when converting from signal intensity to contrast agent concentration (see Perfusion Analysis). This scaling factor will affect the actual pixel values in the perfusion maps when deconvolution is applied. Typically this value should be set so that the resulting perfusion and blood volume values in normal  tissue are according to expected values. This scaling factor must be specified since the exact relation between 1/T1, 1/T2 or 1/T2* and contrast agent concentration is not known in dynamic MRI and depends on factors like contrast agent relaxivity, vascular structure, tissue density and haematocrit. Note that even when AIF deconvolution is applied pixel units are still in arbitrary units since the required conversion factors are not assumed to be known. If the scaling factor is correctly set according to calibration in normal subjects and known perfusion / blood volume values then the pixels values may be converted to mL/100 g  and mL/100g/min respectively for CBV and CBF images. In CT-perfusion analysis the pixel intensity is directly proportional to contrast agent concentration and the scaling factor is then usually set to a value of 1.

Advanced Options

Open Advanced Options to access additional perfusion settings.


Important note on SVD deconvolution:
It should be noted that the actual perfusion values (rBF, rMTT, Delay) obtained are very dependent on the selected or determined SVD threshold, as shown in the figure below. Although automatic (iterative) methods for threshold determination are implemented in nordicICE, there is no guarantee that the resulting perfusion values are clinically meaningful since many unknown factors will influence the quality of the analysis.

The figure above shows the effect of varying the SVD threshold using the ‘truncated SVD’ option with manually defined SVD threshold. Both a too high cutoff (0.7, bottom CBF image) and a too low threshold (0.01, top CBF image) result in erroneous perfusion maps. The middle row shows the resulting residue functions for the same ROI in all three cases. The default SVD cutoff used in the Perfusion Module (0.2) has been set to provide reasonable perfusion estimates with the typical signal-noise ratio obtained in a clinical situation (center images). There is, however, no guarantee that this value will be optimal for all conditions.

Related topics:

Advanced Perfusion Options