A real-time regional adaptive exposure method for saving dose-area product in x-ray fluoroscopy

Abstract

Purpose: Reduction of radiation dose in x-ray imaging has been recognized as a high priority in the medical community. Here the authors show that a regional adaptive exposure method can reduce dose-area product (DAP) in x-ray fluoroscopy. The authors' method is particularly geared toward providing dose savings for the pediatric population.

Methods: The scanning beam digital x-ray system uses a large-area x-ray source with 8000 focal spots in combination with a small photon-counting detector. An imaging frame is obtained by acquiring and reconstructing up to 8000 detector images, each viewing only a small portion of the patient. Regional adaptive exposure was implemented by varying the exposure of the detector images depending on the local opacity of the object. A family of phantoms ranging in size from infant to obese adult was imaged in anteroposterior view with and without adaptive exposure. The DAP delivered to each phantom was measured in each case, and noise performance was compared by generating noise arrays to represent regional noise in the images. These noise arrays were generated by dividing the image into regions of about 6 mm(2), calculating the relative noise in each region, and placing the relative noise value of each region in a one-dimensional array (noise array) sorted from highest to lowest. Dose-area product savings were calculated as the difference between the ratio of DAP with adaptive exposure to DAP without adaptive exposure. The authors modified this value by a correction factor that matches the noise arrays where relative noise is the highest to report a final dose-area product savings.

Results: The average dose-area product saving across the phantom family was (42 ± 8)% with the highest dose-area product saving in the child-sized phantom (50%) and the lowest in the phantom mimicking an obese adult (23%).

Conclusions: Phantom measurements indicate that a regional adaptive exposure method can produce large DAP savings without compromising the noise performance in the image regions with highest noise.

Figure 1

In a conventional system the image is formed by the projection of a single focal spot x-ray tube onto a large-area detector (left). In the SBDX system an image is formed by combining the projections of the multiple focal spots on a large-area x-ray tube onto a small detector (right).

Figure 2

(a) Reconstructed image of the child-sized phantom acquired without adaptive exposure. (b) The detector-count matrix is calculated by integrating the detector counts for each collimator hole. The white outline in the count matrix image indicates the central region that was used to calculate the threshold. (c) The plot shows the integrated detector counts as function of the percentile (only taking collimator holes within the central region into account). The raw threshold is the value at the 30th percentile. (d) The plot shows how the rescan-dependent thresholds are related to the raw threshold. Lower and upper thresholds that are used in the hysteresis algorithm are also shown.

Figure 3

Phantom family used in this study. The black phantom has an iodinated pig heart and is here shown with the additional fat layer (obese adult).

Figure 4

Steps to calculate the noise array. The figure shows images for the child-sized phantom with standard acquisition (top row) and regional adaptive exposure (bottom row). (a) Reconstruction of one frame of the 100 frames taken for each mode. (b) The subtraction image is derived by subtracting two subsequent frames. For clarity we show here the normalized subtraction image that was derived by dividing by the sum of the two frames. (c) The regional noise matrix is derived by dividing the subtraction image into a grid of 15 × 15 pixel regions. For each region the mean is calculated and placed as an element in the noise matrix. (d) The noise array is derived from the noise matrix. The array consists of elements of the noise matrix that are sorted in descending order.

Figure 5

Summary of the pediatric cases studied. For every case four plots are shown. Reconstructed image at isocenter without adaptive exposure (upper left) and with adaptive exposure (upper right). The white circles show the location of lowest performing regions (below the 5th percentile). Plot of the sorted noise array for the experiments with and without adaptive exposure (lower left). In addition the corrected (including regions below the 20th percentile) curve is shown. Exposure map used in the adaptive exposure experiments (lower right). White indicates eight rescans and black indicates 0 rescans; increasing gray levels indicate decreasing numbers of rescans.

Figure 6

Summary of the adult cases studied. For every case four plots are shown. Reconstructed image at isocenter without adaptive exposure (upper left) and with adaptive exposure (upper right). The white circles show the location of lowest performing regions (below the 5th percentile). Plot of the sorted noise array for the experiments with and without adaptive exposure (lower left). In addition the corrected (including regions below the 20th percentile) curve is shown. Exposure map used in the adaptive exposure experiments (lower right). White indicates eight rescans and black indicates 0 rescans; increasing gray levels indicate decreasing numbers of rescans.

Figure 7

Comparison of dose savings for the eight cases studied. Black indicates the measured dose saving and the white and gray bars show the dose saving after the data were corrected to include regions below the 10th, 20th, and 30th percentiles.

Figure 8

Dose savings corrected at the 20th percentile level for different reconstruction heights (at isocenter and ±5 cm from isocenter).