An evaluation of an energy independent CT reconstruction algorithm for use in radiotherapy treatment planning

Abstract

Objective

Radiotherapy treatment planning relies upon density information provided by CT for accurate dose calculations. Hounsfield units (HUs) are converted to electron/physical density via an energy dependant calibration curve. Multiple curves are required to make full use of the available accelerating potentials (kVp). The curves are bi-linear with a discontinuity occurring at soft-tissue densities. The commercial algorithm, DirectDensity^TM (Siemens Healthcare GmbH), constructs a single calibration curve covering all available kVp. This enables the optimisation of the CT image quality, e.g. in terms of contrast, or the reduction of the imaging dose, whilst rendering the radiotherapy treatment dose calculation robust to the energy used to acquire the CT image. We report our investigations on the clinical utilisation of the DirectDensity^TM algorithm for radiotherapy treatments, by using all accelerating potentials, i.e. from 70 kVp up to 140 kVp, available at our CT treatment simulator, in contrast to previous studies that were limited to accelerating potentials spanning a subset of the available kVp.

Methods

The DirectDensity^TM (DD) reconstruction algorithm available on a SOMATOM go.Open Pro CT scanner (Siemens Healthineers) was evaluated using the RayStation v. 9 treatment planning system (RaySearch Laboratories, Stockholm, Sweden) and a CIRS Model 002LFC IMRT Thorax Phantom (SunNuclear, Melbourne, FL), which was imaged at all available kVp with clinical protocols corresponding to various anatomical sites. The DD images were compared to those with the standard reconstruction algorithm acquired only at 120 kVp, as per our routine clinical practice. The effect of increasing kVp on HU is investigated for relevant tissue substitutes. In addition, a dosimetric comparison is performed for a VMAT plan technique with 6 MV X-rays using retrospective patient CT data sets representing four anatomical sites (pelvis, thorax, brain and “head and neck”) with five patients for each site. The original dose distributions were calculated on images acquired at 120 kVp using the standard clinical iterative reconstruction (Qr40) and compared with dose distributions recalculated on images reconstructed with the new DD (Sm40) algorithm.

Results

The maximum difference for radiotherapy doses calculated using images of the phantom reconstructed with Qr40 (120 kVp) or DD (all available kVp) was 0.73%. The patient plans on the anatomically representative sites studied here showed a mean PTV dose difference of −0.2% (s.d. 0.7) for D99%, −0.4% (s.d. 0.4%) for D50% and −0.3% (s.d. 0.4%) for D2%. Incidentally, we found a previously unreported decrease in HU, mostly notable for bone type inserts (~34 HU (cortical bone)), at 110 kVp for the DD reconstructed images. The effect was not noted for the standard Qr40 reconstructions.

Conclusion

DD has a minimal dosimetric impact in the dose calculations for radiotherapy treatments and could be implemented with existing clinical workflows. Attention should be paid to the HU values for images acquired at 110 kVp (DD algorithm), which warrants further investigation.

Advances in knowledge

This is the first paper where DD was evaluated at all available kVp, leading to the incidental discovery of abnormal HU values at 110 kVp for this algorithm.

Introduction

CT is the primary method of imaging used in radiotherapy for the purpose of treatment planning due to its excellent geometric accuracy, soft tissue contrast and the ability to provide with electron density information necessary to calculate dose. ¹ The direct relationship between the CT image grayscale values (Hounsfeld units (HUs)) and electron/physical density is a quantification of the energy dependant X-ray linear attenuation by the human body tissues. The HU-density calibration curve (“HU-d”) is bi-linear with a discontinuity at soft tissue type densities and a strong dependence by the X-ray energy for tissues with higher density. ^2,3 Imaging patients at optimal image quality necessitates different kVp values for different body habitus (or body sizes) necessitating the creation of multiple “HN-d” calibration curves for the radiotherapy dose calculation in the treatment planning system (TPS). This increases the burden of commissioning and QA (for both scanner and TPS) and the risk of selecting an incorrect HU-d curve during treatment planning that does not correspond to the kVp used to acquire the CT image. The latter would result in an erroneous dose calculation estimated to be up to 2%. ^4–6 For these reasons, the typical established clinical practice is to acquire CT images for the planning of RT treatments at a fixed energy (usually 120 kVp for adult patients) and therefore using only one HU-d curve in the TPS, thus restricting the optimisation of the image quality, in terms of contrast- and signal-to-noise ratio, for each individual patient or the ability to reduce the concomitant imaging dose for a given image quality.

Siemens Healthcare GmbH provide an image reconstruction algorithm known as DirectDensity^TM (DD) which reconstructs raw CT projection data into CT values that directly relate to either relative electron or mass densities, whilst removing the energy dependence associated with conventional HU. ^5–8 As such, DD allows the use of a single HU-d calibration curve in the treatment planning system, thereby removing the risk for potential dosimetric errors mentioned previously. ⁸

Our study evaluates the DD reconstruction algorithm for mass density (denoted as Sm40) compared with the standard iterative reconstruction used typically in routine clinical practice (denoted as Qr40). DD’s effect on the HU values for tissues covering the range of densities found clinically is investigated for the first time by using all the available energies (kVp) of a CT scanner. The dosimetric consequences of implementing the DD (Sm40) algorithm at various accelerating potentials (kVp) are assessed and compared to our clinical practice that uses Qr40 with image acquisition at 120 kVp for adult patients.

Methods

Generating the calibration curve

The Gammex Advanced Electron Density Phantom (Sun Nuclear Corporation, Melbourne, FL) was used to characterise the CT numbers (Hounsfield units, HUs) produced when using both the standard iterative reconstruction (Qr40) and the DD (Sm40) algorithms. The Sm40 algorithm reconstructs the CT numbers in terms of mass density rather than electron density, the former being the tissue characterisation metric utilised in the TPS. The phantom has inserts made of different tissue equivalent materials, with certified densities ranging from 0.299 g/cm³ (LN-300 lung) to 1.925 g/cm³ (cortical bone), thus covering the clinical range of human body tissues.

The phantom was scanned at eight different X-ray tube potentials (70, 80, 90, 100, 110, 120, 130 and 140 kVp) in a SOMATOM go.Open Pro (Siemens Healthcare GmbH, Germany) 64-slice CT scanner, with dose modulation switched on, akin to our clinical practice. For each kVp, volumetric (3D) CT images were reconstructed with both the Qr40 reconstruction (as per our standard clinical practice) and the Sm40 algorithm. CT number (HU) to mass density curves (“HU-d”) were generated for all kVp for both reconstruction methods. The mean HU for each insert in the phantom was calculated using a 2 cm² circular region of interest centred on each insert. Because HU depend on the reconstruction algorithm, those calculated with the Qr40 are denoted as HU_standard, whereas those with the Sm40 algorithm are denoted HU_DDm. The HU_DDm for the Sm40 algorithm correlates with mass density (ρ) as described by Eq. (1) below:

$${\displaystyle \rho =\frac{H{U}_{DDm}}{1000}+1}$$ (1)

The “HU-d” calibration curves were input into the TPS for dose calculation (RayStation v. 9, RaySearch Laboratories, Stockholm, Sweden), with the minimum HU of −999 assigned a mass density of 0.001 g/cm³ and the max HU of 3999 to 5.0 g/cm³ as recommended by SIEMENS. ⁸

The influence of the DD algorithm on the HU_DDm of the individual inserts with kVp was investigated. This was done using the same method as above where a 2 cm² circular region of interest centred on each insert was used to measure the mean HU_DDm.

Validation of the calibration curves

The DD_m algorithm was evaluated using a Gammex RMI 467 tissue characterisation density phantom, ⁹ which was not used for the production of the “HU-d” calibration. This allowed for an independent validation of the “HU-d” curve. The RMI phantom was scanned using the clinical abdomen protocol at all eight kVp, reconstructed using the Sm40 (physical density) algorithm and imported to the TPS. 2 cm² regions of interest were drawn on each insert on the phantom images to measure the local HU and thus deriving the corresponding density (g/cm³). The inserts have known calibrated densities, certified from the manufacturer, making it possible to compare the measured densities in the TPS against the certified densities for each insert.

Dosimetric comparison

The CIRS Model 002LFC IMRT Thorax Phantom (SunNuclear, Melbourne, FL) was scanned at all kVp. CT images were reconstructed using both the standard (Qr40) and the DD (Sm40) algorithm. The scans were transferred to the TPS. A virtual mediastinal tumour was delineated and a radiotherapy treatment subsequently planned with 6 MV “flattened” X-rays beams, delivered via a single arc of a Truebeam^TM (Varian Medical Systems Inc, USA) linear accelerator (linac). The dose was calculated using the collapsed cone algorithm and the clinically validated beam model for the linac, as per our clinical treatments. The clinically commissioned “HU-d” calibration for 120 kVp X-rays was used for the plan calculated on the Qr40 CT images. The plan parameters, i.e. modality (X-rays), energy (6 MV FF beams), arc length, collimator and couch angles, were copied onto the DD (Sm40) CT images. One DD generated “HU-d” calibration curve was used to calculate radiotherapy treatment doses for plans using the Sm40 CT images, independent of the energy (kVp) used to acquire the CT scan. A comparison was made between radiotherapy dose distributions calculated either using the Qr40 (120 kVp) images or those reconstructed with the DD algorithm (Sm40).

In addition, patient data of four tumour sites, corresponding to pelvis, brain, head and neck and lung, were investigated. For each site, CT scans of five patients previously treated with radiotherapy at our centre were reconstructed with DD and had their plans recalculated for dosimetric comparison using the same TPS and linac beam model described above.

Results

Calibrations

Figure 1 shows the HU-d CT calibration curve at various energies produced with the Qr40 reconstruction algorithm as per standard practice. In this figure an energy-specific HU to density curve calibration curve has not been applied (the data were acquired directly from the scanner). As expected the calibration line exhibits a high dependence on accelerating potential (kVp) especially for higher density materials, as well as a discontinuity for densities higher than approximately 1.1 g/cm³. In order to clinically implement the use of multiple kVp for imaging, multiple calibration curves (one for each energy) would have to be used resulting in increased risks to accurate dosimetry through choice of incorrect calibration curve and increase in resources through the need to perform routine quality assurance on all the curves. The corresponding HU-d curve created using DD is shown in Figure 2a.

Figure 1.

The HU-d calibration curve for a go.Open Pro CT scanner measured using the “standard” (Qr40) reconstruction algorithm. HU, Hounsfield unit.

Figure 2.

(a) The DirectDensity calibration curve at all accelerating potentials used in this study. The points correspond to the mean HU_DDm for the 2.0 cm² region of interest. (b) The residual differences between the HUDDm and HUstandard and the average HU for each kVp for each density insert. HU, Hounsfield unit.

The DD HU-d (Figure 2a) appears to be linear and less energy-dependent even for higher density materials, which is confirmed by performing a linear regression fit (shown in the figure by the dashed line) that produced a goodness-of-fit (R²) of 0.9971. To further investigate the energy dependence of the apparent X-ray attenuation by various materials when using DD, the relative change of the HU of the phantom’s inserts was plotted at different CT energies (Figure 2b). This shows the difference of the HU for each insert measured at a CT image for a given energy (kVp) and the mean HU for this insert, as calculated for all energies for both the DD reconstruction and the standard Qr40 reconstruction.

For both reconstruction algorithms, the biggest HU variation with energy was most pronounced for the bone type, although this was less pronounced with the DD algorithm. For lung type inserts, the DD algorithm showed increased HU variation over the standard reconstruction. The cortical bone insert had the largest HU_DDm range of 59 over the energies (excluding 110 kVp) which corresponds to a density change of 0.059 g/cm³; or 3.1% compared to the nominal density value. When 110 kVp is included, the range of HU_DDm increased to 81 HU_DDm corresponding to a density change of 0.081 g.cm³; 4.2% cf. the nominal. For comparison, when the Qr40 reconstruction was used, the range of HU_standard over all the energies was 875 HU_standard.

In contrast, the low density insert, LN-300, had a range of 9 HU_DDm when using the DD (Sm40) reconstruction, which corresponds to a density change of <0.009 g/cm³; a 3% change relative to the density of the insert’s quoted value. The Qr40 standard reconstruction gave a variation of 7 HU_standard for this insert. The effect on HU of using one energy to acquire the image data is significantly reduced for soft tissue and lung materials compared with bone type materials. There is no clinical problem with using the Qr40 algorithm for multiple kVp (as long as multiple HU-d calibration curves are used) and modern TPSs can now accommodate this, whereas historically they were limited to single calibration curves. However, each energy requires its own energy-specific curve and as previously mentioned, this introduces a risk in the treatment planning–dosimetry chain arising from incorrect choice of calibration curve. Another factor to consider is the increased resources (not insignificant in a busy department) (staffing, testing/maintenance time) needed to perform routine checking and quality assurance of multiple curves.

Moreover, we found an anomalous feature at 110 kVp for bony type inserts that only occurred on the DD reconstructions. This feature, a sudden drop in the calculated HU, was found to be reproducible on repeat scans with the same scanner settings, when using either the advanced electron density phantom or the RMI electron density phantom. The result was reported to the manufacturer, who also independently reproduced it with their equipment.

Validation

The measured HU_DDm values using the RMI electron density phantom were converted to their respective physical densities using equation 1. The absolute difference (g/cm³) between the density value calculated for each insert and the corresponding one quoted at the phantom’s certificate was derived and presented in Table 1. In general, the calculated mass densities were found smaller compared with the corresponding values quoted by the manufacturer except for the bone type inserts. Over all energies, the mean difference between the calculated minus the quoted densities ranged from −0.07 g/cm³ (acrylic) to 0.04 g/cm³ (cortical bone). The largest difference was found for the cortical bone insert at 70 kVp where the measured density was 0.09 g/cm³ greater than the quoted density.

Table 1.

The absolute difference between the physical densities (ρ_m ) calculated from the measured HU_DDm and the corresponding density quoted by the manufacturer of the RMI phantom, for each of its inserts

INSERT	Quoted mass density g/cm3	70 kVp Sm40			80 kVp Sm40			90 kVp Sm40			100 kVp Sm40			110kVp Sm40			120kVp Sm40			130kVp Sm40			140kVp Sm40			Mean difference from quoted	standard deviation
		HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3	HUDDm	rm g/cm3	Abs diff g/cm3
LN-300 lung	0.30	−704.25	0.30	0.00	−706.47	0.29	−0.01	−705.64	0.29	−0.01	−705.52	0.29	−0.01	−704.84	0.30	0.00	−706.55	0.29	−0.01	−708.21	0.29	−0.01	−704.99	0.30	0.00	−0.01	0.001
LN-450 lung	0.45	−558.52	0.44	−0.01	−561.62	0.44	−0.01	−560.62	0.44	−0.01	−560.52	0.44	−0.01	−562.41	0.44	−0.01	−562.56	0.44	−0.01	−562.89	0.44	−0.01	−563.22	0.44	−0.01	−0.01	0.002
Poly-ethylene	0.92	−122.54	0.88	−0.04	−106.43	0.89	−0.03	−96.80	0.90	−0.02	−92.17	0.91	−0.01	−86.78	0.91	−0.01	−81.92	0.92	0.00	−78.01	0.92	0.00	−73.68	0.93	0.01	−0.01	0.016
Adipose	0.92	−112.59	0.89	−0.03	−103.63	0.90	−0.02	−97.36	0.90	−0.02	−93.61	0.91	−0.01	−89.85	0.91	−0.01	−87.24	0.91	−0.01	−85.04	0.91	−0.01	−82.31	0.92	0.00	−0.01	0.010
Breast	0.99	−56.93	0.94	−0.05	−51.87	0.95	−0.04	−48.41	0.95	−0.04	−47.51	0.95	−0.04	−44.35	0.96	−0.03	−42.71	0.96	−0.03	−41.27	0.96	−0.03	−40.15	0.96	−0.03	−0.04	0.006
True water	1.00	7.34	1.01	0.01	5.42	1.01	0.01	4.86	1.00	0.00	4.45	1.00	0.00	4.34	1.00	0.00	3.74	1.00	0.00	3.55	1.00	0.00	3.22	1.00	0.00	0.00	0.001
Solid water	1.02	8.86	1.01	−0.01	7.59	1.01	−0.01	5.64	1.01	−0.01	3.45	1.00	−0.01	2.18	1.00	−0.01	2.27	1.00	−0.01	0.58	1.00	−0.01	0.40	1.00	−0.01	−0.01	0.003
Brain	1.05	−6.07	0.99	−0.05	4.71	1.00	−0.04	10.39	1.01	−0.03	16.11	1.02	−0.03	18.55	1.02	−0.03	23.45	1.02	−0.02	24.93	1.02	−0.02	28.36	1.03	−0.02	−0.03	0.012
Liver	1.08	99.90	1.10	0.02	97.66	1.10	0.02	93.86	1.09	0.01	92.09	1.09	0.01	91.00	1.09	0.01	88.64	1.09	0.01	87.37	1.09	0.01	86.84	1.09	0.01	0.01	0.005
Inner bone	1.12	143.44	1.14	0.02	141.68	1.14	0.02	141.03	1.14	0.02	139.24	1.14	0.02	134.42	1.13	0.01	138.55	1.14	0.02	137.77	1.14	0.02	136.66	1.14	0.02	0.02	0.003
B200 Bone	1.15	146.56	1.15	0.00	146.3	1.15	0.00	146.57	1.15	0.00	146.20	1.15	0.00	141.25	1.14	0.00	145.46	1.15	0.00	146.58	1.15	0.00	145.11	1.15	0.00	0.00	0.002
C84 resin	1.15	67.63	1.07	−0.08	75.32	1.08	−0.07	83.66	1.08	−0.07	88.53	1.09	−0.06	92.80	1.09	−0.06	92.53	1.09	−0.06	93.17	1.09	−0.06	93.12	1.09	−0.06	−0.06	0.010
CaCO3 10%	1.17	128.71	1.13	−0.04	122.74	1.12	−0.05	122.10	1.12	−0.05	119.21	1.12	−0.05	120.53	1.12	−0.05	122.59	1.12	−0.05	124.91	1.12	−0.05	125.97	1.13	−0.04	−0.05	0.003
Acrylic	1.18	98.92	1.10	−0.08	108.68	1.11	−0.07	111.31	1.11	−0.07	112.21	1.11	−0.07	112.64	1.11	−0.07	105.87	1.11	−0.07	105.54	1.11	−0.07	106.85	1.11	−0.07	−0.07	0.005
CaCO3 30%	1.34	281.30	1.28	−0.06	286.95	1.29	−0.05	290.29	1.29	−0.05	295.33	1.30	−0.04	286.82	1.29	−0.05	299.23	1.30	−0.04	304.42	1.30	−0.04	305.04	1.31	−0.03	−0.05	0.009
CaCO3 50%	1.56	583.90	1.58	0.02	578.34	1.58	0.02	572.84	1.57	0.01	570.66	1.57	0.01	550.96	1.55	−0.01	571.47	1.57	0.01	572.58	1.57	0.01	569.56	1.57	0.01	0.01	0.009
Cortical bone	1.84	931.50	1.93	0.09	899.45	1.90	0.06	886.89	1.89	0.05	876.94	1.88	0.04	844.66	1.84	0.00	870.42	1.87	0.03	868.59	1.87	0.03	863.44	1.86	0.02	0.04	0.026

HU, Hounsfield unit.

Dosimetric comparison

When comparing dose distributions for radiotherapy treatment plans with the CIRS phantom, the mean overall dose differences over all CT energies (using Sm40) from the standard plan (Qr40) for the PTV was: D99% = −0.57% (s.d. 0.07), D50% = −0.09% (s.d. 0.03) and D2% = −0.14% (s.d. 0.02). The largest dose difference of −0.73% for the PTV was found at 70 kVp for the D99% parameter (Figure 3).

Figure 3.

The mediastinal VMAT plan calculated for 6 MV using: (top plot) the standard Qr40 HU_standard curve and an image acquired at 120 kVp, (middle plot) the Sm40 HU_DDm curve and an image acquired at 70 kVp, and (bottom plot) a dose difference map between the two previous plots. HU, Hounsfield unit; VMAT, volumetric modulated arc therapy.

For the patient plans, the dose differences to the PTV ranged from −0.95% (pelvis D50%) to 2.71% (brain D99%) (Figure 4). The doses to the PTV calculated using the Sm40 algorithm were generally lower than the standard Qr40 doses for all the anatomical sites except brain and the D98% and D99% Head and Neck. The doses calculated on the anatomical sites gave an overall mean PTV dose difference of −0.2% (s.d. 0.7) for D99%, −0.4% (s.d. 0.4 %) for D50% and −0.3% (s.d. 0.4 %) for D2%. The largest differences were seen in the pelvis where dose differences of <1% were noted for D50%.

Figure 4.

Box and whisker plots for each anatomical site showing the dose differences between the Sm40 reconstructions compared to the standard Qr40 reconstruction.

Discussion

Dose distribution calculations in the TPS are highly influenced by the the conversion of the CT data into density information, performed via a HU-d, which is usually specific to the energy used for the CT scans. Consequently, routine checking of the CT numbers (HU) and their conversion to density in the TPS is mandated by radiotherapy quality assurance programmes. ^10–12 IAEA recommends a QC tolerance level of ±20 HU, which corresponds to a relative electron density (and by implication a relative physical density) variation of ± 0.02 g/cm³ . ¹³ In our investigations reported here, we found density differences that varied for lung type materials to be less than 0.01 g/cm³ and bone type materials <0.06 g/cm³, when using the DD algorithm.

Initial work by van der Heyden et al, ¹⁴ described an evaluation of the DD algorithm on the Siemens SOMATOM Confidence^® RT Pro scanner (Siemens Healthineers, GmbH). ¹⁴ Their study was limited in that it did not include all the accelerating potentials available on the scanner (they excluded 110 kVp) and is now related to a scanner that has been commercially replaced by the Siemens go.Open series. Further work by D’Alessio et al, ¹⁵ looking at five accelerating potentials (70, 80, 100, 120 and 140 kVp) on the SOMATOM Confidence RT Pro scanner found differences in doses to PTV of <0.3% and OARs < 0.6%. They suggested that although the algorithm was useful for dose calculation, it came with a detriment in image contrast making the DirectDensity images unsuitable for contouring when contrast media was used. The dose differences seen by D’Alessio et al, are comparable with those found in this study, although we found slightly higher dose differences in the pelvis approaching 1%.Flatten et al, ^{15 16} and Feliciani et al, ¹⁷ also reported their implementation of the DD algorithm however both limited their investigations to accelerating potentials of 80, 100, 120 and 140 kV and mainly focused on the artefact reduction algorithms in conjunction with DD. ^16,17 Comparisons of the linear regression fit seen in Figure 2 with similar work by Flatten show the slope of the calibration curve is comparable (both 0.001). The difference in the intercept between this study and that of Flatten was 0.002 HU_DD. The slight difference is attributed to the range of densities the calibration curve is generated over, which defines the range of the fit. Flatten et al, limited their calibration curve to relative densities of <1.7, whereas we allowed ours to extend to densities of up to 1.93. Flatten et al and Feliciani et al demonstrated that there is significant deviation for metal inserts and therefore DD is not applicable for these materials. ^16,17 Similarly, the use of contrast is not recommended for use with DD. In practice, the patient would be scanned using the standard Qr40 kernel at optimal kVp for image contrast and the dosimetry would be performed on the DD reconstruction where the influence of the accelerating potential is removed.

Our evaluation of the dependence of HU_DDm on kVp found negligible difference at low densities, e.g. lung. However there was a stronger energy dependence for bone type densities (Figures 3 and 4), making the difference in density that corresponds to the HU_DDm and the physical one to range between 5 and 1% at 70 and 140 kVp respectively. In addition, there was a decrease in HU_DDm for bone type inserts at 110 kVp, which has not been reported previously. A comparison of the measured HU_DDm compared with those reported by the manufacturer at all the kVp found that the mean difference ranged from −0.07 to 0.04 g/cm³ (-6.1%– 2.1% of the quoted density respectively). ¹⁸ only reported a difference of −0.4% ± 1.7%. However, this was over all the inserts (and excluded 110 kVp) rather than being broken down by insert composition. ¹⁸ For comparative purposes, our study found a difference of −1.4% ± 4.7% (including the 110 kVp data).

The dosimetric impact of the HU-density curve accuracy has been the subject of numerous studies. Tiefling et al, ⁴ found that retrospective calculation of patient plans (treated with cranial radiotherapy) because of the application of incorrect calibration curves, resulted in dose errors of up to 1.5%, associated with variations of up to 20% in HU values, for high density materials such as bone. ⁴ Similarly, Cozzi et al ⁵ and Kilby et al ⁶ reported maximum dosimetric errors of up to 2% where the accelerating potential was the leading contributor to these errors. For our study involving the patient data sets, the dose differences to the PTV ranged from −0.95% (pelvis D50%) to 2.71% (brain D99%). Van der Heyden et al ¹⁴ reported that the relative differences in mean doses and V95% for the sites they investigated were smaller than 1%; however, they were comparing filtered backprojection and DD and with limited accelerating potentials (which would have adjusted their calibration curve linear fit slightly), whereas we compared iterative reconstruction and DD and included both 70 and 110 kV in our analysis. Their study also used dual energy spiral scanning, whereas we limited our study to single energy spiral scans.

Conclusions

The DirectDensity algorithm provides a CT image reconstruction method, which reduces the variation of the HU to changes in the accelerating potential (kVp) used to acquire CT images. As such, it allows for one HU-d in the TPS, irrespective of the energy used in the CT scan, which simplifies QC and reduces the dosimetric errors. However, attention should be paid when using 110 kVp, to the HU of bone type materials.