Streaking artifacts on computed tomography (CT) images are caused by high density materials such as hip prosthesis, surgical clips and dental fillings. The artifacts can lead to compromised clinical outcome due to the inability to differentiate tumor volume and the uncertainties in dose calculation. The goals of our study are to evaluate how GE’s smart metal artifact reduction (MAR) algorithm impacts image quality on phantoms and dosimetry on head and neck patients with dental fillings and pelvic patients with hip prosthesis. Treatment plans calculated on the MAR and non-MAR datasets with the same beam arrangements and fluence are compared. Dose differences between the MAR and non-MAR datasets are not significant. However, substantial reductions of metal artifacts are observed when MAR algorithm is applied. Planning on the MAR dataset is recommended since it improves image quality and CT number accuracy. It also negates the need to contour the artifacts and override the density which can be time consuming.
High density objects, such as hip prosthesis, dental fillings or surgical clips can cause metal artifacts in computed tomography (CT) images. Streaking artifacts are caused by a combination of beam hardening, scatter, noise, photon starvation and exponential edge-gradient effect [
Numerous metal artifact correction methods (MAR) have been proposed since the 1980’s and they have shown to be effective in improving image quality [
For external beam radiation therapy, metal artifacts can compromise a patient treatment in two different ways. First, the streaking artifacts can obscure anatomical details and make target and organs at risk (OAR) delineation challenging. Second, the artifacts change the CT Hounsfield unit and impact the accuracy of dose calculation in a treatment plan. Some studies [
We had the opportunity to assess GE’s smart MAR algorithm on our CT scanner. It uses an automated, three-stage projection based process to improve the image quality [
Our phantom study was conducted with the Catphan® 504 phantom to evaluate the impact of MAR algorithm on CT number sensitometry, geometric accuracy, MTF, low contrast resolution and uniformity. A helical scan was acquired with GE Optima 580 RT-16 CT scanner (GE Healthcare, Milwaukee, WI) with the following parameters: 120 kV, auto mA, 1 s rotation time, 16 × 0.625 mm2 collimation, 2.5 mm slice thickness, 0.938 pitch and 25 cm sFOV. The second CT dataset was reconstructed with the MAR algorithm. Both MAR and non-MAR scans were analyzed with Image Owl QA software (Image Owl Inc., Greenwich, NY).
An in-house manufactured 20 cm diameter cylindrical water phantom was utilized to assess the accuracy of the CT number. The phantom contains three holes in which a 19.0 mm diameter cylindrical stainless steel insert can be positioned in any location while the other holes are filled with cylindrical acrylic inserts. The holes were located in the center of the phantom, in the periphery of the phantom and in between these two locations. These spots were chosen to evaluate how the position of the metal impacts the CT number accuracy. Three scans were acquired, one for each location of the metal insert. The scanning parameters include helical scan, 120 kV, auto mA, 1 s rotation time, 16 × 0.625 mm2 collimation, 2.5 mm slice thickness, 0.938 pitch, 25 cm sFOV, with and without MAR correction. An additional water phantom scan was acquired as the baseline image with the same scanning parameters but with the three holes filled with acrylic inserts. The accuracy of the CT number at six various positions was evaluated with a square ROI in Eclipse TPS™ (version 11.0.31, Varian Medical System, Palo Alto, CA). We compared the CT number from the baseline image without stainless steel to the MAR corrected scans with the stainless steel insert. The dimension of the stainless steel insert was also measured on the CT image by identifying the metal pixel using a threshold HU value (half the maximum metal HU value) [
A total of fifteen H/N cancer patients with dental fillings and ten pelvic cancer patients with hip prosthesis who previously received radiation therapy at our cli- nic were selected for the study after obtaining ethics approval. The study population for H/N cases consisted of 11 male and 4 female with a mean age of 63.9 ± 15.4 years (range 34 - 85 years). For pelvic cases, there were 7 males and 3 females with a mean age of 73 ± 5.0 years (range 65 - 81 years). These patients underwent CT scanning with the following scanning parameters: helical scan, 120 kV, auto mA, 1 s rotation time, 16 × 0.625 mm2 collimation, 2.5 mm slice thickness, 0.938 pitch and 50 cm sFOV. Two CT datasets were reconstructed from the scan, a MAR dataset and a non-MAR dataset. Both datasets were exported to Eclipse TPS™ and delineation of target and organs at risk (OAR) was performed on the MAR dataset by the radiation oncologist and radiation therapist. Clinical plans were optimized and calculated with AAA (version 11.0.31) on the MAR dataset until PTV and OARs met our institution’s clinical dose constraints. Dose calculation grid of 2.5 mm was utilized with the heterogeneity correction. For H/N cancer patients, 6 MV IMRT was the default planning technique with prescription ranging from 45 Gy in 25 fractions to 70 Gy in 35 fractions. For patients with hip prosthesis, either 6 MV IMRT or VMAT was utilized. Prescription ranged from 45 Gy in 25 fractions to 74 Gy in 34 fractions.
Patient # | Age | Gender | Site | Prescription | Treatment technique |
---|---|---|---|---|---|
1 | 79 | Male | Prostate | 74 Gy/37 fractions | IMRT |
2 | 74 | Male | Prostate | 64 Gy/32 fractions | IMRT |
3 | 72 | Female | Anal canal | 54 Gy/30 fractions | VMAT |
4 | 81 | Female | Vulva | 54 Gy/25 fractions | IMRT |
5 | 77 | Male | Prostate | 74 Gy/37 fractions | IMRT |
6 | 70 | Male | Prostate | 66 Gy/33 fractions | VMAT |
7 | 72 | Female | Endometrium | 45 Gy/25 fractions | IMRT |
8 | 65 | Male | Prostate | 66 Gy/33 fractions | IMRT |
9 | 67 | Male | Prostate | 46 Gy/23 fractions | VMAT |
10 | 73 | Male | Rectum | 45 Gy/25 fractions | VMAT |
Patient # | Age | Gender | Site | Prescription | Treatment technique |
---|---|---|---|---|---|
1 | 76 | Male | Base of skull | 45 Gy/25 fractions | IMRT |
2 | 85 | Male | Tongue | 60 Gy/30 fractions | IMRT |
3 | 40 | Female | Hypopharynx | 70 Gy/35 fractions | IMRT |
4 | 53 | Male | Oral cavity | 60 Gy/25 fractions | IMRT |
5 | 75 | Male | Oral cavity | 60 Gy/30 fractions | IMRT |
6 | 52 | Male | Neck | 60 Gy/30 fractions | IMRT |
7 | 34 | Female | Lip | 60 Gy/30 fractions | IMRT |
8 | 79 | Male | Neck | 60 Gy/30 fractions | IMRT |
9 | 56 | Male | Tongue | 60 Gy/30 fractions | IMRT |
10 | 58 | Male | Nasopharynx | 70 Gy/35 fractions | IMRT |
11 | 75 | Male | Hypopharynx | 70 Gy/35 fractions | IMRT |
12 | 84 | Female | Neck | 70 Gy/35 fractions | IMRT |
13 | 60 | Female | Nasopharynx | 70 Gy/35 fractions | IMRT |
14 | 69 | Male | Oral cavity | 60 Gy/30 fractions | IMRT |
15 | 62 | Male | Oral cavity | 50 Gy/25 fractions | IMRT |
prosthesis. For the H/N cases, no special attention was paid to avoid treatment beams entering through the dental fillings because these regions were small.
After the treatment plan was approved by the radiation oncologist, contours from the MAR dataset were copied onto the non-MAR dataset. Next, a separate dose calculation was performed on the non-MAR dataset with the same treatment field arrangement and fluence as the clinical plan. Dose differences between the two CT datasets were evaluated for PTV and OARs. Some patients had multiple PTVs but only results from the high dose PTV will be presented here. In this study, none of the metal artifacts were contoured with density over-rides.
To quantify the percentage and absolute difference between MAR and non- MAR plans, the following conventions were utilized:
For target volume evaluation, the conformity index was utilized. This is a ratio of prescription isodose volume to the target’s volume. Endpoints for PTV include D99% (dose to 99% of target volume) and V100% (volume receiving prescription dose). For H/N OARs, we compared the mean dose to the parotids and ma- ximum dose to spinal cord and brainstem. For pelvic plans, we assessed the DVH of bladder, rectum, femoral head, iliac crest and the genitalia.
Comparisons between scans with and without MAR algorithm on the Catphan phantom demonstrate similar results for image quality. Geometric accuracy, MTF, CT number for various materials and low contrast resolution were very similar, if not identical. There was a small difference for noise level.
Evaluation of CT number at six various locations of the in-house phantom was conducted on the central axis slice.
Image quality tests | MAR | No. MAR | |
---|---|---|---|
CT number sensitometry | Air, expected = −1000 HU | −964 HU | −965 HU |
PMP, expected = −200 HU | −182 HU | −183 HU | |
LDPE, expected = −100 HU | −92 HU | −93 HU | |
Polystyrene, expected = −35 HU | −39 HU | −39 HU | |
Acrylic, expected = 120 HU | 118 HU | 119 HU | |
Delrin, expected = 340 HU | 334 HU | 334 HU | |
Teflon, expected = −990 HU | 917 HU | 917 HU | |
Geometric accuracy | Measured distance, expected = 0.5 mm | 0.49 mm | 0.49 mm |
MTF | Critical frequency, 50% (cycles/cm) | 4.01 | 4.02 |
Critical frequency, 10% (cycles/cm) | 6.78 | 6.80 | |
Critical frequency, 5% (cycles/cm) | 7.45 | 7.45 | |
Critical frequency, 2% (cycles/cm) | 8.11 | 8.07 | |
Low contrast | Details at 1% contrast | 5 mm | 4 mm |
Details at 0.5% contrast | 8 mm | 7 mm | |
Details at 0.3% contrast | 9 mm | 9 mm | |
Uniformity | Mean CT value | 7.7 HU | 7.6 HU |
Noise | 15.6 HU | 13.2 HU |
the MAR algorithm does not alter the CT number when there is no high density material. In
ference was smaller without MAR algorithm when the metal insert was at ROI position 3. This is contrary to what we observe for other ROIs.
The physical diameter of the stainless steel was compared to the measurement from the CT image which over-estimated the insert by 0.9 mm. The MAR algorithm appears to correctly reconstruct the dimension of the stainless steel insert.
Similarly to our phantom study, we see a significant reduction of metal artifacts with our clinical CT datasets when the MAR algorithm is applied. However, re-
sidual artifacts are still present.
For all fifteen H/N patients, the average percentage differences in conformity index, D99% and V100% are −0.3% ± 0.9%, −0.1% ± 0.1% and −0.1% ± 0.5% respectively. For all ten pelvic patients, the average percentage discrepancies in conformity index, D99% and V100% are −8.8% ± 11.4%, −0.1% ± 0.4% and −8.8% ± 12.1% respectively.
Patient #13 in
Patient #4 in
For H/N OARs, we compared the mean dose to the parotids and maximum dose to the spinal cord and brainstem as shown in
For the pelvic cases, we performed a plan subtraction in Eclipse between plans calculated on MAR and non-MAR datasets. An example is shown in
with a double hip replacement. Isodose levels corresponding to ±2% of the prescription are shown in orange and magenta. Absolute dose differences larger than 2% are near the boundary regions of hip prosthesis and skin surface. A review of all ten of our pelvic cases indicates the dosimetric changes between calculations performed on a MAR and a non-MAR datasets are not significant. This finding is similar to the study from Li et al. [
Remarkable efforts have been made in the recent years in developing commercial algorithms to reduce metal artifacts and noise in CT images. In this paper, we provided an experimental and clinical evaluation of one commercially available MAR algorithm for CT simulations in radiation therapy. We found GE’s smart MAR algorithm to be effective in reducing artifacts for H/N patients with dental fillings and pelvic patients with hip prosthesis. The reduction of streaking artifacts allows radiation oncologists to accurately delineate targets and organs at risk. This negates the need to increase target margin which may lead to more normal tissue toxicity. Furthermore, the accuracy of CT number is improved when MAR algorithm is applied. GE’s software is able to correctly characterize the dimension of the stainless steel insert in our phantom study. Although the algorithm provides an improved image dataset, there are still some residual artifacts in the corrected images. Han et al. [
The degree of dose discrepancy between treatment plans calculated on a MAR dataset and a non-MAR dataset depends on a few factors. Our study shows dosimetric impact from hip prosthesis is greater than dental fillings because hip prosthesis produces more artifacts. The proximity of the organ to the high density material is crucial as well. A larger dose difference is observed when the organ of interest is closer to the high density material. The beam arrangement can also play a role as more uncertainties are introduced when a field is going through a high density material. Dose differences between the plans can be positive or negative depending on the type of metal artifacts. Dark streaks have lower HU and can introduce hot spots whereas bright streaks have higher HU and introduce cold spots. Our findings conclude there is minimal dosimetric difference between treatment plans calculated on the MAR and non-MAR datasets. This is supported by the studies from Li et al. [
In our study, we chose to compare plans calculated on the MAR dataset versus the non-MAR dataset. We did not compare MAR plan to non-MAR plan without heterogeneity correction because the variation between these two plans includes differences from the heterogeneity correction. Since the focus of our investigation is on the metal artifacts, we did not want to include the dosimetric effects due to heterogeneity. One weakness of our study is that we do not know the composition of the hip prosthesis and dental fillings. Thus we are unable to correlate the dosimetric impact based on the type of the metal.
One limitation with GE’s smart MAR algorithm is that the sFOV must be smaller or equal to 50 cm. At our clinic, when the patient’s anatomy extends out- side of 50 cm sFOV, target and OAR contouring is performed on the MAR dataset. These contours are copied onto the non-MAR dataset for dose calculation purpose. In addition, metal artifacts on the non-MAR scan need to be contoured with 0 HU assigned.
This study indicates GE’s smart MAR algorithm can improve CT number accuracy and correctly characterize the dimension of the metal insert without impacting the overall image quality. However, residual metal artifacts are still observed in the MAR corrected images. The dose differences between IMRT and VMAT plans calculated on the MAR and non-MAR datasets depend on the proximity of the organ to the high density material, the type of streaking artifacts and the beam arrangement of the treatment plan. With our study population of 15 H/N patients with dental fillings and 10 pelvic patients with hip prosthesis, we found the dosimetric difference to be minimal between MAR and non-MAR datasets for both PTV and OARs. There are several advantages of planning on the MAR corrected images. First, there is substantial reduction of metal artifacts which can allow the radiation oncologist to contour targets and OAR more accurately. Second, treatment planning time can be reduced because there is no need to contour the artifacts and override the density. Last, the MAR corrected images will provide better reference images for image guidance. Therefore, MAR corrected images are recommended for radiotherapy treatment planning.
Huang, V.W. and Kohli, K. (2017) Evaluation of New Com- mercially Available Metal Artifact Reduc- tion (MAR) Algorithm on Both Image Qua- lity and Relative Dosimetry for Patients with Hip Prosthesis or Dental Fillings. Interna- tional Journal of Medical Physics, Clinical Engineering and Radiation Oncology, 6, 124-138. https://doi.org/10.4236/ijmpcero.2017.62012