Validation of CT brain perfusion methods using a realistic dynamic head phantom
(a) The segmentations of a 5 mm slice of the head phantom (partial volume and resolution effects excluded to clarify the features). (b) An example of a 5 mm noise image used in the digital head phantom (acquired from a scan of a physical head phantom at 150 mAs) with nonuniform noise distribution inside the skull. (White overlay added to noise mask here to highlight the position of the skull.) (c) A frame from the head phantom with contrast enhancement beginning to appear in the arteries and veins. (d) 3D cerebral artery volume seen from the front of the head.
Tissue residue functions are scaled by the respective CBF values, the level of the plateau represents the CBF. Shown here are tissue residue functions modeled to represent gray tissue, white tissue, diseased gray tissue, and diseased white tissue. The longer MTT of the diseased tissue is evident by the broader functions. The relationship between CBF, MTT, and CBV (the area under the curve) is defined by the CVP. (Note: Blood flow is defined as milliliters of blood per 100 ml of tissue per minute.)
An example of the ROI’s drawn on white tissue and gray tissue on a time maximum-intensity image of clinical CTP data. These ROI’s were saved and transferred to the CTP parameter maps where the ROI statistics were used to compare typical clinical results with the hybrid head phantom.
The TAC of the clinically acquired AIF (rescaled using the AUC of the VOF) and VOF. The gray and white matter TAC’s are generated using Eq. (2). These curves represent the attenuation enhancement before the effects of partial volume and noise. Tissue time attenuation curves have been multiplied by 10 for clarity.
Scatter plot showing correlation between input CBF (defined by generated TAC) and calculated CBF (output from perfusion software) when using bcSVD, sSVD, and FBD. Noise-free head phantom: Δ, with clinical dose (150 mAs): □, and with high dose (305 mAs): ◊. The solid line represents an ideal correlation, data points above this line are overestimated, and data points under the line are underestimated.
Correlation between CBF (in healthy tissue defined by a radiologist) calculated from clinical data and the result calculated from the head phantom, error bars represent variability of the mean CBF in clinical results (1 standard deviation). Also show are clinical and head phantom CBF values showing how the mean CBF values are changed by altered bcSVD calculation parameters.
Comparison of variance (standard deviation) in clinical CBF maps and head phantom CBF maps.
Effects of varying the strength of the Gaussian filter (left) and the oscillation index (right) in the bcSVD based calculations on the head phantom. Dotted lines define real CBF values for comparison and are labeled on the far right.
CBF diagnostic image quality improvement using the head phantom to alter bcSVD. (a) Clinical data CBF map with default pma parameters. (b) Clinical data CBF map after optimization. (c) Head phantom CBF map with default pma parameters. (d) Head phantom CBF map after optimization.
MTT and CBV maps showing good agreement between the results from the clinically acquired data and the head phantom. (Vessels have been excluded.) (a) Clinical MTT map, (b) phantom MTT map, (c) clinical CBV map, and (d) phantom CBV map.
Values used to define tissue residue function parameters for the various tissue types. These values were chosen from the mean values defined in Ref. 25.
Results of linear fit to mean measured CBF vs. real CBF for the three software packages using the hybrid head phantom at different dose levels. The ideal result is a slope of 1 and a Y-intercept of 0. All linear fits showed a good R 2 correlation (=0.92), indicating that the results are linear, but with a bias dependant on the dose.
Sensitivity of perfusion algorithms to delayed arrival time of blood in diseased tissue. All values are CBF in ml/100 ml/min. (Noise-free phantom only.)
Article metrics loading...
Full text loading...