Assessment of portal image resolution improvement using an external aluminum target and polystyrene electron filter

In a preliminary study, an 8 mm external aluminum target produced the largest proportion of useful low-energy photons for the 6 MeV electron beam than thick-optimized low-Z targets, such as 18 mm beryllium, 14 mm carbon, and 4 mm titanium [10]. The Al-target beam produced a total photon yield of 64.2% and a low-energy photon percentage among the total photons of 34.4%. This result was similar to that observed in a previous study, in which a low-Z target material was inserted in the linac head carousel, regarding the percentage of photons in the 25–150 keV range [2].

However, 35% of the incident primary electrons remained in the Al-target beam. The electrons transmitted through the target caused noise in the portal images and increased patient exposure. Therefore, these remaining electrons should be removed. A polystyrene plate has been used as a filter to remove the electrons in previous studies, because polystyrene has a higher stopping power and lower photon beam generation efficiency than those of the higher Z materials for the electron energies of interest [2, 7, 16]. In this study, the GEANT4 Monte Carlo simulation was used to determine the optimal polystyrene filter thickness to maximize electron removal. As shown in Fig. 1, the polystyrene filter was located immediately below the aluminum target. The percentages of the remaining electrons, photons, and low-energy photons were analyzed by changing the polystyrene filter thickness from 4 to 20 mm. A phase space file provided by Varian was used as the beam source for the 6 MeV electrons, and an International Atomic Energy Agency (IAEA) format phase space file was generated using the GEANT4 code with the geometric input parameters of a medical linac [17‐19]. The IAEA format phase space was recorded above the upper jaws, and the phase space was used as the beam source. The polystyrene filter and other linac head components, such as the upper and lower jaws, Mylar window, and 8 mm aluminum target, were modeled and simulated in the GEANT4 code. The field size was set at 20 × 20 cm² at the isocenter. As shown in the Fig.1, the scoring plane was located immediately below the polystyrene filter, and the energy spectra of the particles were recorded and analyzed. The number of histories in the space file of the 6 MeV incident electrons was 7 × 10⁹, and the statistical uncertainty of the doses calculated by the Monte Carlo simulation was within 2%. The cut-off range of the photons, electrons, and positrons was 0.01 mm.

To validate the accuracy of the Al-target beam modeled by the Monte Carlo simulation, the percentage depth dose (PDD) and lateral profiles were compared with the actual data obtained using the customized aluminum target and polystyrene filter. The measurements were performed using the Blue Phantom (IBA Dosimetry, Germany) and Farmer-type ion chamber with a volume of 0.64 cm³ (EXRADIN A12, Standing Imaging, USA). The PDD for the aluminum target beam was acquired for a field size of 20 × 20 cm² at a source surface distance (SSD) of 100 cm in water. The lateral profile was measured at a maximum dose depth and field size of 20 × 20 cm². The field sizes were set using linac jaws. To obtain the dose distribution by Monte Carlo simulation, a virtual water phantom was created using the same settings as in the actual measurement. The overall size of the water phantom was 50 × 50 × 50 cm³, and the voxel size was 5 × 5 × 5 mm³. The cut-off range of the photons, electrons, and positrons was 0.1 mm. The gamma factor analysis was used to evaluate the difference between the modelled and measured profiles. The measurement data set was used as the reference dose distribution, and the calculated data set is used as the evaluated dose distribution. The linear interpolation of evaluated dose distribution was conducted as 1 mm pixel (data point) spacing. The acceptance criteria of the gamma factor were based on the dose-difference of 3% and distance-to-agreement of 3 mm, respectively.

All images acquired with the two imaging beams were imported to epidSoft software provided with the phantom for the quantitative analysis of the image contrast resolutions. All phantom images presented in this study were inverted to assess the image sharpness. No gray scale manipulation was performed.

Table 3 lists the Monte Carlo simulation calculated proportions of the low-energy (25–150 keV) photons and electrons generated by the 8 mm aluminum target and 6 MeV electron beam with respect to the polystyrene thickness. The ratio of the remaining incident electron beam gradually decreased as the polystyrene thickness increased, and the electrons were nearly eliminated at the polystyrene thickness of 20 mm with minimal effect on the ratio of low-energy photons. This result is shown in Fig. 2, where the low-energy photon spectra for the aluminum target beams with various polystyrene thicknesses were almost identical, except in the very low-energy region. Furthermore, these results confirmed the high stopping power and low photon generation efficiency of the polystyrene plate. In previous studies, polystyrene electron filters of various thicknesses were used to remove the remaining incident electrons, because low-Z targets placed in the linac head carousel are not thick enough to stop all primary electrons [2, 7, 16, 22]. Low-Z target material types and thicknesses are difficult to determine owing to the continuous slowing down approximation (CSDA) range to a required tolerance within the confined space of the carousel. Even if a low-Z target is manufactured and installed in accordance with the carousel specification, a polystyrene filter is required. Conversely, the external aluminum target with a polystyrene plate used in this study requires simple machining and contains minimal installation constraints. Based on the results of the simulation, a 20 mm thick polystyrene plate was chosen as the electron filter, customized and placed under the 8 mm aluminum target.

To validate the Monte Carlo simulation, in terms of the Al-target beam quality, the modeled and measured PDDs and profiles in water were compared. Figure 3(a) shows the modeled and measured PDD curves of the Al-target beam for a field size of 20 × 20 cm² and an SSD of 100 cm. The PDD curves were normalized to the maximum dose. The Monte Carlo simulation results of the PDD of the Al-target beam were consistent with the measured data to within 1.6%. Figure 3(b) shows lateral profiles for modeled and measured data for the Al-target beams at d_max, for a field size of 20 × 20 cm², at an SSD of 100 cm. The lateral profile curves were normalized to the dose values on the central axis. Figure 3(b) inset shows the corresponding gamma factor analysis for the modeled and measured profiles. For the criteria of 3% and 3 mm, all gamma values in the open beam region were less than one.

Figure 4 shows the MTF curves for the two imaging beams. The Al-target beam produced better MTF values at each spatial frequency than those of the 6 MV photon beam. The critical spatial frequency (f₅₀) was also higher for the Al-target beam with a value of 0.745 lp/mm than those for the 6 MV photon beam with a value of 0.451 lp/mm. As shown in Fig. 5, all diagonal line pair images for the two imaging beams were clearly visible, and the lines and gaps were easily differentiated. As shown in Fig. 4, for the two imaging beams, the MTF values for the spatial frequency from 0.125 to 0.33 lp/mm were higher than the critical spatial resolution (f₅₀) value. Figure 6 shows the horizontal line pair images of the two imaging beams for the spatial frequencies from 0.5 to 3.3 lp/mm. The Al-target beam produced darker and clearer line pair images than those of the 6 MV photon beam. For the 6 MV photon beam, the line pairs of the images were increasingly blurred as the spatial frequency increased, and distinction was not possible from a frequency of 2.0 lp/mm, where the MTF value fell below the limiting spatial resolution (f₁₀) value. Conversely, the line pairs of the images for the Al-target beam remained distinguishable at 2.0 lp/mm, and the MTF value at this frequency was higher than the limiting spatial resolution value. These results suggested that the MTF of the detector system was dependent on the photon energy [16]. Thus, Al-target beams containing a large low-energy photon component are capable of producing higher MTF values than those of 6 MV photon beams. Figure 6 also shows the line pair images at the spatial frequencies in the range of 2.5 to 3.33 lp/mm for the two imaging beams, the Al-target beam produced the clear images; however, the lines and gaps were not distinguishable. This result was supported by the MTF curve shown in Fig. 4, where the MTF values for each spatial frequency in the range were lower than the limiting spatial resolution.

Previous studies using line-pair-type phantoms have reported that low-Z target beams could improve the spatial resolution compared to standard 6 MV photon beams [13, 16, 23]. Roberts et al. characterized spatial resolution using a full thickness 2 cm carbon target and a 4 MeV electron beam and found, using QC3 phantom line pair images, that an improvement in MTF was observed for the custom target beam. Connell and Robar investigated the effects of different beam generation parameters, such as the target atomic number, target thickness, and incident electron energy, on the spatial resolution and demonstrated that low-Z targets could produce high-resolution images as good as those acquired with 6 MV photon beams. However, in these previous studies, modifications of the linac head compositions and modulations of the incident electron energy were required, and thus, their findings are not easily applied to existing equipment. Conversely, in this study, the installation of external targets was simple because no modification of the linac head configuration was required, and thus, these results will be more applicable to the existing linacs if a detector suitable for the Al-target beams is designed.

Contrast resolution demonstrates the ability to detect subtle differences in grayscale in acquired images [21]. Figure 7 shows the contrast-detail test element images of the two imaging beams and also shows the relative contrast differences between the holes. Different hole depths and diameters indicate contrast and detail, respectively. The contrast-detail distributions are shown as color bars, where the green bars represent holes that can be distinguished with contrast differences ≥5%. The yellow bars represent barely distinguishable holes with contrast differences < 5% and ≥ 3%, and the red bars represent indistinguishable holes with contrast differences < 3%. The Al-target beam showed 11 green bars, 5 yellow bars, and 11 red bars. The 6 MV photon beam showed 8 green bars, 5 yellow bars, and 14 red bars. In the image of the Al-target beam, each hole was distinguishable with minimal noise or blurring, indicating that the Al-target beam resolved the smaller and lower-contrast objects better than that of 6 MV photon beams. This was attributed to the different attenuation coefficients of the Al-target beam and 6 MV photon beams, as shown by eq. (8). These findings show that the Al-target beams can improve the low-contrast resolution of the fine structures that cause small grayscale differences compared to the 6 MV photon beams.

The next stage of this work involves the assessment of the image qualities produced using the Al-target beams with different object thicknesses. Because the low-energy component of the Al-target beams may be diminished when imaging thicker objects, a qualitative and quantitative analysis is required. A specific detector for the Al-target beams will be designed. However, because most MV portal imagers attached to the linac are optimized for MV photon beams, the Al-target beams cannot be used for imaging, and thus, detector design will be required.