banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
Investigation into the optimal linear time-invariant lag correction for radar artifact removal
Rent this article for


Image of FIG. 1.
FIG. 1.

The normalized measured and ideal responses of the Varian 4030CB a-Si FP to 600 frames of x-ray exposure. (125 kVp, 32 mA, 17 ms, 15 fps). A 4% signal increase is seen in the measured RSRF. After the x-rays are turned off, a FSRF with 2.5% first frame lag that slowly decays away is observed. Note the break in the y-axis.

Image of FIG. 2.
FIG. 2.

Reconstructions of a simulated, uniform pelvic phantom (42 × 26 cm) with (a) no lag and (b) simulated detector lag. The mean signal difference between the two marked ROIs is 45 HU. (c) Sinogram with no lag subtracted from the sinogram with lag. A vertical blurring is seen at the edges of the object (indicated with arrows) in the difference sinogram. Detector elements in section 1 experience a large dynamic range while those in section 2 are always in object shadow. (d) CT reconstruction of a pelvic phantom showing the radar artifact. Scanned at 125 kVp, 80 mA, 30 ms on a Varian 4030CB table-top system. Reconstructed with filtered backprojection with a voxel size of 1 mm2 × 5 mm. Between the specified ROIs there is a mean difference of 51 HU. Window, level = 200, 0 HU.

Image of FIG. 3.
FIG. 3.

1 × 1 × 5 mm reconstructions of a pelvic phantom (30 cm × 22 cm) collected with the detector in (a) dynamic gain and (b) low gain (4 pF) mode. Window, level = 250, 0 HU.

Image of FIG. 4.
FIG. 4.

Residual error in fitting multiexponential models of order N to measured step-response data for two different exposure intensities. Low exposure equals 3.4% and high exposure equals 84% of a-Si FP saturation.

Image of FIG. 5.
FIG. 5.

(a) Raw (detector counts) and (b) normalized rising and falling step responses for several different exposure intensities. In (a), the measured lag increases as the exposure intensity increases. In (b), the measured data were normalized by the exposure intensity immediately before x-ray turn-off. As the exposure intensity increases, the relative amount of measured lag decreases.

Image of FIG. 6.
FIG. 6.

The normalized RSRF data for each frame rate (5, 10, and 30 fps) plotted as a function of (a) frame number and (b) time. In (a), the RSRF decreases with increasing frame rate and in (b), the four data sets overlay each other. (c) The raw FSRF data for each frame rate scaled by the incident exposure at 5 fps and plotted versus frame number. (d) i trap(t) for the four frame rates. The charge per unit of time is constant across frame rates, indicating that the detrapping process is a continuous function of time.

Image of FIG. 7.
FIG. 7.

(a) The 600th frame in an RSRF measurement with no gain correction. (b) Raw lag frame two with no gain correction applied; the greatest lag variation occurs near the edges of the detector. Gain-corrected and normalized lag images for lag frames (c) 2 and (d) 100. (e) Histograms of pixel values for raw lag frame 2 and gain-corrected lag frames 2, 50, and 100. For all images, the object in the lower right corner is a c-Si photodiode used to normalize the x-ray source output. Exposure is 3.7% of saturation.

Image of FIG. 8.
FIG. 8.

Uncorrected, RSRF corrected, and FSRF corrected data sets are shown. 1st and 50th frame lags for the uncorrected data are 2.9 and 0.49%. The correction done with the RSRF calibrated IRF shows a very flat corrected RSRF but a significantly overcorrected FSRF. The 1st and 50th frame lags become −0.54% and −0.25%. Conversely, the FSRF-based correction has a residual exponential rise in the corrected RSRF, but the 1st and 50th frame lags are much reduced to 0.24 and 0.017%.

Image of FIG. 9.
FIG. 9.

Measured and corrected (a) RSRF and (b) FSRF data at a low and high exposures (3.4 and 84% of a-Si FP saturation). The IRFs used for data correction were calibrated from (a) 3.4% RSRF and (b) 3.4% FSRF data. The corrected 84% RSRF data are overcorrected, while the corrected 3.4% RSRF data is nearly flat. For the corrected 3.4% FSRF, 1st frame and 50th frame lag is reduced from 3.6 and 1.0% to 0.12 and 0.038%. For the corrected 84% FSRF, 1st and 50th frame lag is overcorrected from 2.3 and 0.42% to −1.1 and −0.54%.

Image of FIG. 10.
FIG. 10.

The uncorrected and corrected (a) RSRFs and (b) FSRFs for 32 × 32 pixel ROIs on the FP. The IRFs used for correction were derived from the (a) RSRF and (b) FSRF data of the central ROI. The corrected central ROI, the average corrected ROI, and the corrected ROI that is maximally deviant from the average are shown. For the average of all ROIs, the 1st and 50th frame lags are reduced from 4.0 and 0.62% to 0.66 and 0.1%. For the maximally deviant ROI, the 1st and 50th frame lags are reduced from 5.2 and 0.94% to 1.9 and 0.42%, which is 4× greater than the average residual 50th frame lag.

Image of FIG. 11.
FIG. 11.

Measured and corrected FSRF data for 10 and 40 s of x-ray exposure at 25% of a-Si FP saturation. IRF was calibrated with the measured 40 s FSRF data. For the corrected 40 s FSRF, 1st and 50th frame lags are reduced from 2.8 and 0.61% to 0.44 and 0.023%. For the corrected 10 s FSRF, 1st and 50th frame lags are reduced from 2.6 and 0.45% to 0.52 and 0.082%.

Image of FIG. 12.
FIG. 12.

(a) The uncorrected reconstruction of the pelvic phantom with the ROI pairs denoted. Reconstructions with the (b) smallest average error and (c) largest maximum error, which correspond to IRFs derived from an FSRF at 3.4% and an RSRF at 1.6%. (d) Reconstruction using an ROI-based calibration. (e) Uncorrected reconstruction of a uniform acrylic head phantom. Reconstruction with the (f) smallest average error (FSRF at 1.6%) and (g) largest error (FSRF at 84%). (h) Reconstruction using an ROI-based calibration. Window, level = 200, 50 HU for the pelvic phantom and 100, 50 HU for the head phantom.

Image of FIG. 13.
FIG. 13.

The maximum and average absolute errors between ROIs within a pair for the pelvic and head phantoms. The CBCT projection data were corrected with IRFs derived from different exposure intensities using either the FSRF data in (a) and (c) or the RSRF data in (b) and (d). For the RSRF data, the 0.5% exposure did not give a suitable fit for the multiexponential model.

Image of FIG. 14.
FIG. 14.

Difference images between the uncorrected pelvic phantom and the best LTI corrected data sets of (a) the CBCT reconstructions and (b) the raw x-ray detector sinograms. (b) Looks different than the simulation in because of the offset geometry acquisition. For the sinogram, projection angle and time increase in the downward direction. Window, level for (a) is 100, 0 HU and for (b) is 350, 200 detector counts.


Generic image for table

Percent of a-Si FP saturation signal for different exposure intensities at 125 kVp with 0.5 mm Ti.

Generic image for table

IRF parameters from FSRF data at 3.4% exposure. a n has units of frames−l and b n has units of detector counts with lag/detector counts without lag.


Article metrics loading...


Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Investigation into the optimal linear time-invariant lag correction for radar artifact removal