Computed tomography imaging parameters for inhomogeneity correction in radiation treatment planning.

Modern treatment planning systems provide accurate dosimetry in heterogeneous media (such as a patient' body) with the help of tissue characterization based on computed tomography (CT) number. However, CT number depends on the type of scanner, tube voltage, field of view (FOV), reconstruction algorithm including artifact reduction and processing filters. The impact of these parameters on CT to electron density (ED) conversion had been subject of investigation for treatment planning in various clinical situations. This is usually performed with a tissue characterization phantom with various density plugs acquired with different tube voltages (kilovoltage peak), FOV reconstruction and different scanners to generate CT number to ED tables. This article provides an overview of inhomogeneity correction in the context of CT scanning and a new evaluation tool, difference volume dose-volume histogram (DVH), dV-DVH. It has been concluded that scanner and CT parameters are important for tissue characterizations, but changes in ED are minimal and only pronounced for higher density materials. For lungs, changes in CT number are minimal among scanners and CT parameters. Dosimetric differences for lung and prostate cases are usually insignificant (<2%) in three-dimensional conformal radiation therapy and < 5% for intensity-modulated radiation therapy (IMRT) with CT parameters. It could be concluded that CT number variability is dependent on acquisition parameters, but its dosimetric impact is pronounced only in high-density media and possibly in IMRT. In view of such small dosimetric changes in low-density medium, the acquisition of additional CT data for financially difficult clinics and countries may not be warranted.

Introduction

Treatment planning systems (TPS) have evolved from using actual data to analytical approaches derived from pencil beams.[1] The older generations of TPS provided dosimetry exclusively in water (without inhomogeneity correction) based on regular fields[2] and Clarkson integration[3] for irregular fields for patient treatment. However, a patient’s body is not homogenous and not water equivalent. Rather, it is complex and heterogeneous with natural variation in tissues such as lung, cartilage, bone, and implanted high-density and high-atomic number (Z) materials such as dental fillings, pacemakers, and prostheses.

Attempts have been made to provide correction factors for lung inhomogeneity, beginning with Batho.[4] McDonald et al.[5] provided a comprehensive set of tables for lung correction with respect to energy, field size, and depth. The equivalent tissue-air ratio was introduced to correct for inhomogeneities;[67] this was followed by power law.[89] Use of computed tomography (CT) data did not start until the introduction of the generalized equation based on CT pixel-by-pixel correction.[10] Various other algorithms[1112131415] have been proposed over time including algorithm based on electron transport.[16] A detailed evolution of the inhomogeneity correction and its impact on patient care has been provided by AAPM Report 85.[17]

Inhomogeneity corrections were also debated as clinicians were reluctant to use them without clinical outcome data.[17181920212223] However, inhomogeneity correction has become an essential part of treatment planning in modern therapy and is required for intensity-modulated radiation therapy (IMRT).[24] Recent advances in dose calculation using advanced algorithms based on Monte Carlo modeling such as pencil beam, convolution/superposition, and collapsed cone have facilitated improved dosimetry and dose calculation accuracies.[17252627] However, advanced dose algorithms require electron density (ED) from CT data to account for the effects of inhomogeneity rather than physical density scaling as was advocated by the older algorithms such as equivalent path length (EPL).[28] To correlate the CT numbers in a patient’s CT study with the corresponding ED values, a CT number – ED calibration curve should be determined. The CT number of any voxel is given as below which is represented in Hounsfield units (HU):

Where x, y, z is the coordinate of a voxel, µt and µw are the linear attenuation coefficients of tissue in a voxel (x, y, z) and water, respectively. CT number is a quantity and HU is a unit; however, these terms are interchangeably used. By definition, HU is 0 for water and −1000 for air at standard temperature and pressure. It is obvious that CT number depends on the attenuation property of a medium, and it should be dependent on beam energy, density, and atomic number.[29] It follows that the CT number of a given tissue is not constant. Rather, it depends on tube voltage (kilovoltage peak [kVp]), field of view (FOV), scattering conditions, and vendor-specific CT reconstruction algorithms.

The tissue characterization in terms of CT number and ED calibration used in TPS have been proposed by several investigators[3031] using a commercial phantom. The calibration curve (CT-ED) is stored in the database of the TPS for dose calculation purposes. The CT-ED curve and its impact on dosimetry has been documented in the context of older dose calculation algorithms to be 1.3%, 0.8%, 0.5% for Co-60, 6 MV and 21 MV beam, respectively, and was independent of either EPL or power law calculation.[32] Morgan et al.[33] used advanced TPS and quantified the dosimetric impact very similar to data previously presented by Jones et al.[3435] for variation in lung density and field size.

The selection of CT-scanner and technical consideration for TPS has been provided by Cao et al.[36] Each CT scanner manufacturer optimizes CT images based on the selection of body section to be imaged; however, different techniques may be used depending on the scan protocol. Since the selection of technique on a CT scanner may provide the same tissue with a different CT number, the treatment planner must know the impact of such changes. The variation of CT numbers due to different scanning parameters has been noted by many investigators,[373839] and some studies have been performed to investigate its dosimetric effect by using inhomogeneous cubic or anthropomorphic phantoms.[4041] Most of these studies evaluated the absolute doses per monitor unit (MU) to a single point (such as isocenter or a reference dose point) without consideration of dose coverage to targets and critical organs. The impact of kilovoltage setting for low-Z inhomogeneity for a scanner has also been reported to be insignificant clinically,[42] however variability among different CT scanners studies has been limited. Recently Zurl et al.[43] compared CT parameters and showed that variation up to 20% in HU could be noted; however the impact on dose is limited to only 1.5%. Thus, effect of CT number for photon and electron beam Monte Carlo calculations has been noted to be different and needs attention.[44] Ebert et al.[45] provided variability of CT number from a GE scanner at various kV settings and tube currents. It was shown that tube current (mAs) does not a play role and only kV provides variation in CT number.

The objective of this review article is to evaluate the variation of CT numbers of different scanning parameters such as tube voltage (kVp), and physical and reconstruction FOV on several commercial scanners and compare it with other publications. The dosimetric impact of different CT number - ED calibration from different scanners is also evaluated for clinical cases with emphasis on dose coverage to tumor targets and its impact to critical organs. Conclusions are then made if such CT data are needed within the limit of dosimetric accuracy for radiotherapy centers and countries where additional scan could be a financial hardship to the patients.

Computed Tomography Number to Electron Density Calibration

To revisit CT number-ED calibration, a tissue characterization phantom (RMI, Gammex, Middleton, WI, USA) was used to evaluate under different scanning conditions. The phantom consists of a solid water disk approximating the size of an average pelvis that contains interchangeable rods made of various tissue equivalent materials. The physical density (g/cm3) ranges from 0.3 (LN-300 lung) to 1.84 (cortical bone), and the corresponding ED relative to water varies from 0.292 to 1.707. The RMI CT-phantom is commonly used in radiotherapy clinics in the United States. The quality assurance in the manufacturing of these tissue-equivalent plugs is very precise (<1% variation), which was verified among five phantoms.[46] The phantom was placed in the center of a CT gantry by careful alignment with lasers and scanned with different imaging protocols using various tube voltages (80–140 kVp) on each scanner. Two reconstruction fields of view (33 cm and 48 cm) were chosen to reconstruct the images with a 512 × 512 matrix with 5 mm slice thickness contiguously. After image reconstruction, a circular region of interest (ROI) of 1.5 cm diameter was placed on each density plug and the mean CT numbers of the ROIs were recorded. To minimize the effect of image artifacts and beam hardening, multiple CT scans of the phantom were acquired with different combinations of insert position and the resultant mean CT numbers were averaged. The same process was repeated on several scanners including wide bore (85 cm) and small bore (72 cm) Philips PQ5000 scanners (Philips HealthCare, Andover, MA, USA) and a Somatom 4 scanner (Siemens Medical Solutions, Malvern, PA, USA). The CT number - ED table was generated in each configuration as described by Constantinou et al.[30] The resultant CT number - ED conversions were compared between different scanners, reconstructed FOV, and tube voltages.

Dosimetric Impact of Computed Tomography Number to Electron Density Calibration

The CT number - ED calibration tables were imported into the Eclipse TPS (Varian Medical Systems, Palo Alto, CA, USA) and were used to investigate its impact on dose calculations. Under institutional review board exempt status, two typical cases (lung and prostate) were chosen in this study. Treatment planning was performed using the analytical anisotropic algorithm that provides superior inhomogeneity correction as reported by many investigators.[474849] To investigate the dosimetric impact in low-density tissues, three-dimensional (3D) conformal as well as IMRT plans were generated to achieve optimum coverage of a representative tumor lesion centrally located in the right lung of a patient for both 6 and 15 MV X-rays. In each plan, a different CT number - ED calibration table for a given tube voltage (80 kVp–140 kVp) was used for inhomogeneity correction. The remaining parameters, for example, beam arrangements, and MU were kept the same. The difference in dose coverage of the planning target volume (PTV) and organs at risk (OAR) (lung and heart) were compared by evaluating the dose-volume histograms (DVHs). In the second case, the CT study of a prostate cancer patient was chosen so that some beams passed through the hip with high-density bone compared to the soft tissue. A 3D treatment plan using 4-field box technique was generated as well as a 7-field IMRT plan. The dose differences in PTV and OAR (rectum, bladder, and femoral heads) with various ED tables were evaluated. For comparison in both cases and techniques, 3D conformal radiation therapy (3DCRT) and IMRT, the MU calculated for the 140 kVp CT scan for optimum coverage of the PTV was used for calculation in other CT scans with different kVp setting. Again, the planning parameters, for example, beam arrangements, fields, and MU were kept the same. For clinical evaluation of treatment plans, a new concept based on volume difference from DVH (dV-DVH) is introduced to provide to compare competing DVHs when the differences among the DVHs are negligible. The dV-DVH of a structure is a plot of the difference between the volume of the structure covered by a given dose and a reference volume at the same dose. The dV-DVH magnifies the subtle difference between DVHs that are closed to each other. This proposed concept dV-DVH provides a better evaluation tools for plan comparisons where DVHs have small differences and are not differentiated. The clinical implication of dV-DVH is yet to be realized as we believe this is the 1st time that the concept of dV-DVH is introduced in treatment planning.

Outcome of Computed Tomography Number to Electron Density Calibration

CT number versus relative ED for different tube voltages and reconstructed FOVs were plotted for a Philips PQ5000 and a Siemens Somatom 4 scanner in Figure 1a and b, respectively. The discontinuity (bump) at around density 1.1 is typical of RMI phantom and has been noted by other investigators.[4246] This is probably due to the artifact in the plug that has different chemical compositions but same physical density. The differences in CT numbers versus tube voltages are minimal in the density region from 0.3 (lung) to 1.0 (water). This discrepancy becomes significant for high-density materials and can reach up to 43% for cortical bone (1668 HU at 80 kVp vs. 1167 at 140 kVp) with a trend that higher kVp yields a lower CT number. This is probably due to the increase in photoelectric attenuation for lower photon energies which lead to higher CT number. Full- and half-FOV reconstructions have little effect on the CT numbers of all materials for both scanners; the only exception was the 11% difference (1869.4 HU vs. 1686.4 HU) for cortical bone at 80 kVp for the Somatom 4 CT.

Figure 1

Computed tomography number versus tube voltage for a Philips PQ5000 (a) and Somatom 4 scanner (b). For both scanners, data are shown with full (f) and half (h) field of view. Note that the computed tomography number is relatively unaffected for low-density materials for both kilovoltage peak and field of view

The illustration in Figure 2 compares the CT number to relative ED calibration curves of the two CT scanner vendors for the same FOV. Significant differences in CT number were observed for high-density tissues between the two scanners. Lower kVp tends to have larger discrepancy between scanners with the maximum difference of 15% at 80 kVp. The CT number to ED calibration curves for the Philips PQ5000 scanner with different gantry apertures (72 cm and 85 cm) are compared as shown in Figure 3. Again at low density, there is no difference in CT numbers. However, large differences are noted at higher densities especially for bone. The maximum difference in CT number was 10% occurring at 80 kVp for cortical bone.

Figure 2

Comparison of computed tomography number from two different manufacturers (Philips and Siemens) with same scan aperture and reconstructed field of view. Note significant changes in computed tomography number for bone with two scanners

Figure 3

Comparison of computed tomography number for Philips PQ5000 with different apertures (72 cm and 85 cm). There is a very little difference in computed tomography number between two scanners

Dosimetric Impact in Clinical Cases

The dosimetric impact of ED variation was revisited to evaluated two clinically relevant cases (lung and prostate). For the lung case, the differences between 3DCRT and IMRT were minimal for PTV coverage for all ED tables. Figure 4 shows the DVHs for the 3DCRT plans with 6 MV beams. It can be seen that for a given structure (PTV or OAR), the DVHs are practically indistinguishable for all CT number to ED curves obtained with different kVp. Similar findings were also observed for 15 MV beam (not shown). For the IMRT plans, the small differences among the various plans are probably due to the plan optimization process. Overall, the differences in DVHs caused by different CT number to ED calibrations were negligible (<1%) for all 4 plans (6 MV and 15 MV and 3DCRT and IMRT). This is probably because the CT number variation of lung tissue for different tube voltages has been shown to be minimal [Figures 1 and 2]. To better examine and quantify the small deviation in DVHs, the PTV volume coverage of all the ED calibrations was compared to that of 140 kVp. The differences in the range of dose levels (90–110%) were minimal. For both the 6 MV and 15 MV plans, the calibration of 80 kVp led to the largest deviation from that of 140 kVp with less volume coverage. This might be caused by high-density material presented in the paths of the beams. Nevertheless, the maximum difference was only 1.1% for both plans and can be considered as clinically insignificant. Similarly, no significant difference was found for DVHs of critical organs such as spinal cord, heart, and right lung as shown in Figure 4.

Figure 4

Dose-volume histograms of planning target volume, spinal cord, heart and lung of a tumor lesion centrally located in the right lung calculated using computed tomography number to electron density calibrations of different tube voltages (80–140 kVp)

Compared with the lung case, the DVHs for the prostate PTV demonstrated a slightly larger difference between different tube voltage calibrations as shown in Figure 5a and b and Figure 6a and b for 3DCRT and IMRT, respectively. The dV-DVH concept was introduced to magnify the effect of differences in DVH which is shown in the insets [Figures 5b and 6b] whereas differences in DVH seem to be small. For both 6 MV and 15 MV plans, lower kVp calibration tends to result in less volume coverage for dose range from 95% to 100% of the prescription dose. The largest differences were −9.6% for 6 MV and −8.3% for 15 MV fields, respectively, both occurred at the 97.5% of dose prescription in 3DCRT. This dose deviation can be mainly caused by the presence of large bony structures around prostate and the considerable variation of CT number - ED versus tube voltage of high-density materials as demonstrated in Figures 1 and 2. With regard to critical organs such as rectum and bladder, the tube voltage caused very small variation in dose distribution as shown in Figures 5 and 6.

Figure 5

(a) Comparison of dose-volume histogram for a prostate three-dimensional conformal radiation therapy with various electron densities associated with tube voltages (80–140 kVp) and (b) magnified view of dose-volume histogram for planning target volume only. Also note the plot of dV-dose-volume histogram in inset providing useful information where dose-volume histogram cannot be easily differentiated

Figure 6

(a) Comparison of dose-volume histogram for a prostate intensity-modulated radiation therapy with various electron densities associated with tube voltages (80–140 kVp) and (b) magnified view of dose-volume histogram for planning target volume only. Also note the plot of dV-dose-volume histogram in inset providing useful information where dose-volume histograms cannot be easily differentiated

The differences are slightly higher in IMRT plans as shown in Figure 6b compared to Figure 5b for the 3DCRT. Some of these differences are inherent to IMRT optimization where an exact solution is not achievable and variability in inter- and intra-institution and planner are significant.[50] The differences in 3DCRT and IMRT for the prostate case are <2% and <5%, respectively, based on analysis of dV-DVH as shown in Figures 5b and 6b.

Discussion

Two types of curves CT number versus ED and ED versus CT numbers are shown in various references.[3132404245] However, CT number versus ED curve is better suited as ED is unknown variable which should be evaluated based on scanners derived CT number. We reevaluated and quantified the variation of CT number - ED calibration between different vendors, tube voltages, and FOVs and its impact on radiation treatment planning and dose calculation as shown by other investigators.[324042] After scanning an ED calibration phantom using the same scanning parameters on six different scanners, Constantinou et al.[30] observed more than 200 HU difference in cortical bone between different scanner vendors. By analysis of published data for a number of scanners, Thomas[32] showed that there was no great difference in the relationship between CT number to relative ED for low-density materials between the different manufacturers and calibration techniques. These are confirmed in this study. For high-density materials, considerable differences between data sets from different machines and measurement techniques were observed. Analytic calculation based on effective depth showed that changes in inhomogeneity correction factors were less than 1.5% for a 10% change in CT number. In a similar study, CT number was found to be stable with respect to different acquisition parameters, except for the tube voltage setting that can lead to errors of about 300 HU for high-density materials.[40] The authors also investigated the dosimetric impact using a simplified anthropomorphic phantom with a single bone embedded in a tissue equivalent material and found around 2% maximum error. Guan et al.[41] investigated the dosimetric impact of different CT number - ED curves for full lung plus three typical bone sites under single beam irradiation. The dose per MU was found to be 2% higher for 80 kV than that of 130 kV at a depth just beyond bone for high-density bones. For low-density bones and lung, the difference is only 1% or less for different kV. A recent study by Zurl et al.[43] indicated that even though the CT number variation can be significant, its dosimetric impact is limited to only 1.5% concluded from study based on 28 real patients. Compared with the above studies, we observed similar variation in CT number among different scanners. The tube voltage was found to be the most influential factor, whereas other scanner parameters have minimal effect. We also found that CT number deviations are minimal for low-density materials but become significant for high-density materials. Instead of comparing single point dose or MU/Gy in simplified phantoms, we investigated the impact on dose-volume coverage in real patient plans. We found very small differences for PTV coverage in lung, but relatively higher difference for the prostate case as evaluated using dV-DVH.

As demonstrated in our study, high-density materials may have a large effect on the accuracy of CT number and dose calculation. In additional to bones, contrast agents and metal implants are two high-density materials that are commonly present in patient CT scans. The influence of CT contrast agents on dose calculation had been investigated by Ramm et al.[51] A typical bolus of 3 cm3 and CT number of 1400 HU was found to cause overdose of up to 7.4% and 5.4% for 6 MV and 25 MV photon beams, respectively. It was suggested that contrast agents with CT number lower than 500 HU and volume less than 5 cm in diameter will not cause significant changes (<1–3%) in dose calculation. The situation of metal implant is more complex because it not only causes saturation of the CT number in the metal implant itself, but also generates significant artifacts that affect the accuracy of CT numbers of other materials. It is unfortunate that none of the scanners can provide artifact-free CT data as well as none of the TPS can give accurate dose distribution with high-Z materials.[525354555657585960] In view of such findings, along with the guideline of dosimetric considerations for patients with hip prostheses as provided in AAPM TG 63 report,[56] it is prudent to eliminate beams passing through metals to reduce dosimetric error.

One of the biggest drawbacks in TPS is the estimation of actual CT number which is marred by the artifact. Artifact reduction algorithms are an active area of research in diagnostic imaging for the interpretation on images as well as dosimetry in radiation therapy. These algorithms have limited success as shown in various references.[2853596162636465666768697071727374] An extended CT-scale calibration to 16 bit has been proposed which has been shown with limited success in the prediction of electron densities of metal inserts.[6175] Some TPS provide ad-hoc corrections by inserting electron/physical density up to Z = 22 (4.5 g/cm3) for titanium, however prosthesis such as steel, molybdenum, chromium, and various other alloys are still beyond reach of most TPS. Monte Carlo-based TPS which are on the horizon might prove to be useful in such situations.

For most of the studies reported so far, the dosimetric impact of different CT number to ED conversion was mainly focused on photon beams. The variability in CT number could be large but its impact on dose in low-density medium or for thorax and pelvic malignancies are limited (<2%). In addition, most scanners provide very similar CT numbers, as shown by Cheng et al.[46] The influence of scanning parameters on CT number and corresponding dosimetric impact on dose calculation for electron and proton beams require further investigation which has not been discussed here due to range and stopping power issues.[76]

Summary

Based on previously published papers and revisiting this issue from a separate angle, it is concluded that the variation of CT number versus scanning parameters and CT scanner vendors is different. CT numbers for the same material from different CT scanners are expected to be variable. However, for low-density media, CT number changes are minimal with scanners and X-ray energies but deviations could be significant for high-density materials. A higher tube voltage gives lower CT number, while other parameters such as reconstruction FOV and scanner aperture have little effect on CT number. For low-density tissues, inhomogeneity correction can be successfully (±2%) applied with a single CT - ED table for 120–140 kVp. Larger variation in dose coverage was observed for high-density tissues between different tube voltages. Thus, it may be advisable to perform more strict calibrations corresponding to tube energy especially when IMRT is used. The dV-DVH is a simple and useful tool for dosimetric comparison as it enhances graphically the small differences between the DVHs that are superimposed on each other. Validity of acquiring different CT data for planning should be evaluated based on necessity and actual gain in dosimetry especially for poor patients, centers, and countries.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.