If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
The metal artefact reduction algorithm improved computed tomography number accuracy.
•
Contouring accuracy of critical structures was increased.
•
Clinical contouring time was significantly reduce.
Abstract
Background and purpose
Metallic implants cause artefacts in computed tomography (CT) images and can introduce significant errors to structure visualisation and dosimetric calculation within the radiotherapy planning process. This study evaluated an orthopaedic metal artefact reduction algorithm and its effect on the CT number, image noise, structure delineation, and treatment dose.
Methods
Raw CT data were reconstructed using standard filtered back projection and an artefact reduction algorithm to create ‘standard’ and ‘corrected’ images. A phantom containing tissue-mimicking inserts and two titanium plugs was imaged. The average CT number was compared to baseline data acquired without metal inserts. Data from 11 pelvic external beam radiotherapy (EBRT) patients with bi- or uni-lateral hip implants were retrospectively analysed. The clinically used treatment plans were re-computed on the corrected images. A prostate-mimicking phantom containing metal ‘implants’ was imaged, and 11 observers contoured both reconstructions.
Results
The artefact reduction algorithm improved the CT number in those areas most affected by metal artefacts and decreased noise by 19 % (P =.04) Changes in dose distributions on corrected images compared to those calculated using the current clinical protocol were clinically insignificant. Volumes contoured on the corrected phantom images had larger Dice coefficients than those contoured on the standard images (P =.001), as well as a 36 % lower standard deviation in volumes.
Conclusion
This study demonstrates that the metal artefact reduction software reduces the error in CT numbers, can improve delineation accuracy, and can reduce inter-observer variability. It has the potential to streamline the planning pathway and improve treatment planning accuracy.
Implanted medical devices such as dental fillings and orthopaedic implants can cause artefacts in computed tomography (CT) images. These are introduced via scatter, partial volume effects, aliasing, beam hardening and photon starvation [
]. The result is streaking artefacts where dark bands appear downstream of dense objects, cupping artefacts, where the periphery of objects appears falsely bright and a general reduction in image quality. Artefacts such as these can produce images with incorrect CT numbers (a relative measure of the radio density of an object). Radiotherapy planning using these images can lead to inaccuracies in the calculated dose distributions [
Artefacts are routinely dealt with in clinic by contouring the artefact and metal implant and manually adjusting the relative electron density values. Low density artefacts are assigned to water density, whereas the implant volumes are assigned to a density appropriate for that material. This process is time consuming and can introduce uncertainty in dose distribution.
There are various ways in which the effects of metal artefacts can be reduced, once of the most common being artefact reduction algorithms. There are multiple artefact reduction algorithms available which broadly fall into four categories (iterative, interpolation, filtering, and hybrid combinations), with many techniques specific to the CT manufacturer. To varying degrees these algorithms have been shown to reduce metal artefacts, improve the CT number accuracy, reduce noise in CT images, improve critical structure visualisation, aid target delineation, and improve dose calculation [
Reduction of artefacts caused by orthopedic hardware in the spine in spectral detector CT examinations using virtual monoenergetic image reconstructions and metal-artefact-reduction algorithms.
The primary objective of this study was to assess a commercially available reconstruction algorithm and quantify its impact on CT number accuracy, image noise, structure delineation accuracy, and calculated dose distribution for patients with orthopaedic hip implants undergoing pelvis radiotherapy.
2. Methods
2.1 Ethics statement
This study was designed to assess current care and assessed patient data retrospectively. The study protocol did not demand changes to treatment/ patient care from accepted standards for any of the patients involved, and the findings of this study were not generalizable. All patient data was anonymised before use and no data were shared outside LTHT (Leeds Teaching Hospitals Trust). In line with the health and research authority, 2017, this study was deemed a service evaluation and did not require research ethics committee review, health research authority approval, or a confidentiality advisory group application. In addition, in-house ethics approval covered all radiotherapy related patient data used in this work.
2.2 Patient selection
This study included patients with bi- or unilateral titanium hip implants (Fig. 1) that had undergone radiotherapy treatment of the pelvis. Fourteen patients were included in the study. Twelve patients had prostate treatments and two had rectum treatments. Four of these patients had bilateral hip implants.
Fig. 12D transverse slices are taken from a planning CT of a patient (#4 in this study) undergoing prostate radiotherapy with bilateral hip implants. Both images are of the same slice, windowing, and levelling settings. Image (a) was reconstructed using the standard algorithm, and the image (b) using the O-MAR algorithm. Considerable streaking and photon starvation artefacts can be seen across the centre of the standard image, almost entirely obscuring the prostate.
A tissue characterisation phantom (Model 467 electron density phantom, Gammex, Middleton, WI, USA) was imaged using a Philips Big Bore CT simulator (Philips Healthcare, Best, NL). All scans of the phantom were performed using the departmental clinical pelvis scanning protocol with the following settings; 16 × 1.5 mm collimation, 0.813 pitch, 0.75 s rotation time, 13 s scan time, 120 kV, 163 mAs, 2 mm slice thickness, 2 mm increments, 300 mm scan length, and a 600 mm field of view. The phantom was positioned at the isocentre of the scanner bore.
The tissue characterisation phantom is constructed of a disc of solid water equivalent material containing twenty holes filled with interchangeable inserts of various tissue and water equivalent materials (Fig. 2). Wax (50 % paraffin wax, 50 % bees wax) inserts were constructed in-house to replace lung-mimicking plugs and air gaps (holes containing no inserts), both of which are a contraindication for the commercially available reconstruction algorithm O-MAR (Orthopedic Metal Artefact Reduction) [
Metal artefact reduction for orthopedic implants (O-MAR). White Paper, Philips CT Clinical Science, Andover, Massachusetts. 2012. Available at: https://www.philips.co.uk/healthcare/education-resources/publications/hotspot/O-MAR-metal-artefact-reduction.
]. The phantom was imaged twice; with and without two grade 5 titanium rods replacing the inserts in positions 3 and 7, mimicking the presence of bilateral hip. Grade 5 titanium was used here as it is commonly employed for medical implants [
Fig. 2(a) CT image (FBP reconstruction) of the tissue characterisation phantom, where a number indicates insert placement. (b) Table identifying which material-mimicking plug is in which position. There is no tabulated density value for the in-house constructed wax insets. Metal inserts were placed in positions 7 and 3.
The raw image data were reconstructed using the standard Philips reconstruction algorithm (filtered back projection) and the O-MAR reconstruction algorithm (which incorporates an iterative projection modification process), resulting in 4 images in total. Each image was analysed using a local Matlab code (The MathWorks Inc, Natick, Massachusetts). The central slice of the CT image was aligned with a template which identifies 20 circular regions of interest (ROIs) with diameter 2.5 cm corresponding to the areas of the phantom inserts. The software extracted the average CT number and standard deviation in each ROI. A two-tailed Wilcoxon signed rank test was used to assess the statistical significance of the difference between the CT numbers in the standard and O-MAR corrected image sets [
The CT numbers measured on the tissue characterisation phantom containing no metal were compared to the baseline results acquired during CT simulator commissioning in 2018.
2.4 Patient contouring study
Using the raw CT data for each of the 14 patients, images were reconstructed using the standard and O-MAR algorithms. Three clinically trained observers contoured the bladder, rectum, and bowel loops on each of the ‘standard’ images, followed by the ‘corrected’ images. One observer also contoured the prostate. All contouring was conducted according to the clinical guidelines performed only on transverse slices. Observers were allowed to vary the thresholding and windowing to identify the necessary structures. Upon completion, each contour was peer reviewed to ensure its clinical suitability. A two-tailed Wilcoxon signed rank test was used to assess the statistical significance of the difference between contours conducted on the standard and corrected images [
A tissue-equivalent ultrasound prostate phantom (CIRS, Norfolk, VA, USA) was used for contouring purposes. The phantom was CT imaged using the aforementioned pelvis scanning protocol, and the CT data reconstructed using the standard algorithm.
The two previously described titanium inserts were positioned inside the water tank, on either side of the phantom, and aligned with the ‘prostate’ to mimic bilateral hip implants. A second CT scan of the phantom was acquired, and the data were reconstructed using the standard and O-MAR algorithms.
Twelve clinically trained observers contoured the prostate, with participants contouring the ‘standard’ image first, followed by the ‘corrected’ image. All contouring was performed according to the clinical guidelines, as stated above. The Dice coefficient, precision, and sensitivity were calculated to assess each of the contours' accuracy compared to a ‘gold standard’ contour conducted on the image containing no metal [
]. A two-tailed Wilcoxon signed rank test was used to assess the statistical significance of the difference between contours conducted on the standard and corrected images [
Three patients from the cohort had been CT imaged but did not continue with their treatment. Consequently, eleven patients were included in the dose distribution study, four of whom had bilateral hip implants.
The raw CT data were reconstructed using both the standard and O-MAR algorithms. The clinically used contours for the artefact area, metal implant, and PTV (planning target volume) were used to analyse both image sets. The contours and the patient’s treatment plan were delineated using the standard image set. As per the clinical protocol, the artefact volume was identified for the standard images using a CT number threshold value and forced to a relative electron density of unity. The metal implant volume was forced to a relative electron density of five. All contours, excluding the artefact volume, were copied onto the corrected images, and the electron density adjusted for the metal implant as above. The clinically used treatment plan (VMAT) was then re-calculated on each image set, and the dose distributions were calculated. A gamma analysis was carried out, and the dose-volume-histogram (DVH) statistics for the PTV were assessed to compare target dose coverage.
2.7 Treatment planning pathway audit
Following the clinical implementation of O-MAR, an independent audit was performed to identify benefits to the clinical workflow. Staff (imaging specialist radiographers, planning staff and clinicians) were asked to comment on the impact of image verification at treatment and the time saving during the planning process. The pathway for ten patients was reviewed.
3. Results
3.1 CT number study
The average CT number difference in each ROI between non-metal images and the baseline was less than 0.35 %. There was no difference in CT number between the ‘standard’ and ‘corrected’ non-metal images.
For areas containing the most severe metal related artefact the O-MAR algorithm improved the CT number accuracy. For example, at position 2, the CT number rose from −5.6 % to −2.3 % compared to the baseline value (a 3.3 % percentage point increase in CT number accuracy). For the other two ROIs most affected by metal-related artefact (18 and 20), the increase in accuracy was 4.6 % points and 4.0 %, respectively. The remaining ROIs reported negligible changes in CT number accuracy (Fig. 3). A statistically significant reduction in mean CT number standard deviation for all ROIs of 19.52 % (P =.04) was found between the ‘standard’ (78.48 [range: 42.41–260.27]) and ‘corrected’ (63.16 [range: 25.70–229.01]) images.
Fig. 3(a) ‘Standard’, and (b) ‘corrected’ CT images of the tissue characterisation phantom with metal inserts in positions 7 and 3 (as indicated in Fig. 2). (c) The absolute difference in HU number in each ROI for both the ‘standard’ and ‘corrected’ images compared to the baseline result. A larger value indicates where the measured HU number was further from the baseline result. The ROIs are numbered according to the template in Fig. 2 and ranked from left to right in increasing electron density. The positions corresponding to the wax and metal have not been included due to a lack of baseline results.
A statistically significant (P =.001) increase in mean prostate volume of 3.5 cm3 was found for prostate delineations on the corrected images compared to the standard images. There was no statistically significant difference in volume for any other contoured organs (bladder, rectum, and bowel loop).
3.3 Phantom contouring study
A mean increase in DICE, precision and sensitivity for the corrected image vs the standard image was seen to be 0.05(P =.001), 0.08(P =.001), and 0.12(P =.0001) respectively. A statistically significant increase in volume of 7 cm3 (P =.003) was found for the corrected image volume (48.5 cm3) compared to the standard image volume (41.5 cm3). The gold standard contour volume was 47.5 cm3. Additionally, there was a reduction in standard deviation for the contoured volume of 1.9 cm3 for the corrected image (3.2 cm3) compared to the standard image (5.1 cm3).
3.4 Dose distribution study
Using criteria of 3 mm and 3 %, the gamma pass rate was consistently greater than 99.8 % when comparing doses calculated on the ‘corrected’ images to those calculated on the ‘standard’ images. With the criteria reduced to 2 mm and 2 %, the pass rate was still over 99 % (Fig. 4). Patients 2, 4, 7, and 11 were those with bilateral hip implants.
Fig. 4Gamma pass rate, comparing dose distributions calculated on O-MAR corrected images and those using the standard clinical protocol. The green line indicates the required gamma pass rate when planning verification at LTHT with a 3 % and 3 mm tolerance. Patient numbers 2, 4, 7, and 11 had a bilateral hip implant. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
There was no statistically significant difference in the PTV DVH statistics (D98, D50, and D2%) calculated on the standard and corrected images. In most cases, the percentage difference was less than 1 %. For one patient, the differences were more prominent than 1 % (1.4, 1.2, and 1.8 % difference for D98, D50, and D2% of the PTV respectively), however the DVH statistics were still within clinical tolerance for treatment [
A mean time saving of 30 min per patient was achieved during the planning process. Staff reported increased confidence in soft tissue matching due to improved image quality. In addition, regions of rectal gas were found to be more easily identifiable.
4. Discussion
A commercially available reconstruction algorithm was investigated by assessing CT number accuracy, image noise, structure delineation accuracy, and calculated dose distribution for patients with orthopaedic hip implants undergoing radiotherapy of the pelvis. The impact of implementing this algorithm on the treatment planning pathway was also assessed.
The CT number study was conducted to identify whether the O-MAR algorithm improved the accuracy of the CT number in CT images affected by metal artefacts. Inserts 18 and 20 were the most affected by streaking artefacts, followed by inserts 2 and 9, as shown in Fig. 4. At these four locations, the O-MAR algorithm significantly improved the CT number accuracy. However, the results show that in other locations, the O-MAR algorithm reduces the CT number accuracy, resulting in CT numbers which lie further away from the baseline (e.g. insert number 10) where the CT number accuracy falls from 0.8 % below the baseline to 4.3 % with O-MAR applied. This contradicts the findings of multiple published studies [
] where no definitive answer is presented regarding the circumstances under which the O-MAR algorithm decreases CT number accuracy. Repeat scans were conducted with the inserts in different positions within the phantom, which demonstrated similar results. One possibility is that the accuracy of the O-MAR algorithm depends on various contributing factors, such as the position of the metal, proximity to the metal/artefact, density of the area, and the location of other high-density materials to the measured area. Additionally, the negatively affected areas could result from artefacts still retained in the corrected images [
Koehler T, Brendel B, Brown KM. A new method for metal artefact reduction. In: The Second International Conference on Image Formation in X-ray Computed Tomography, June 24-27, 2012, Salt Lake City, Utah, USA; authors version; 2012.
The standard deviation of the CT number provides a measure of the noise, typically correlated with image quality. It was found that the O-MAR algorithm significantly reduced the noise in images with metal artefacts. This is in agreement with multiple similar studies [
CT radiation dose reduction in patients with total hip arthroplasties using model-based iterative reconstruction and orthopaedic metal artefact reduction.
Quantitative analysis of orthopedic metal artefact reduction in 64-slice computed tomography scans in large head metal-on-metal total hip replacement, a phantom study.
Analysis of the images identified that the contoured prostate volume was significantly larger on the corrected images than the standard images (mean ∼8 %). Hansen et al. presented a similar result, where contoured volumes were generally larger after reducing the artefacts [
]. The patient's bladder, rectum, and bowel loop volumes showed no significant change between the corrected and standard images. This is likely because the artefacts resulting from orthopaedic hip implants typically affected the prostate more than the other organs mentioned above due to its approximate to the hip. Therefore, the visibility of those organs will not have been as significantly impacted by the O-MAR algorithm as the prostate. This indicates that the algorithm causes no additional changes (e.g. geometric distortions) to areas of the images not affected by the artefact, which supports the safe use of the algorithm in clinical situations.
Prostate contouring was performed by a single clinically trained observer and assessed by peer review. Typically the prostate would be contoured by a clinician; however, this was not possible due to time and resource constraints. The results presented here should therefore be considered preliminary until further work can be conducted with clinicians’ participation.
The change in the contoured prostate volume on the patient images indicates that the O-MAR algorithm can have a marked effect on structure visibility. However, there is no ‘ground truth’ to compare these contours, so their accuracy could not be assessed. In order to investigate this, the contouring of a known structure was performed using a prostate-mimicking phantom. This study showed that the O-MAR corrected images allow for significantly improved contours with a more significant Dice coefficient, precision, sensitivity, and a smaller volumetric difference when compared to the gold standard. These results imply that the O-MAR algorithm can increase the accuracy of contouring on clinical images, which is currently-one of the largest sources of uncertainty in the treatment planning process [
]. Additionally, the O-MAR algorithm can reduce inter-observer variation, as shown by a decrease in the standard deviation of the volumes contoured.
A gamma pass rate consistently above 99.8 % was measured when comparing dose distributions calculated on the images using the standard clinical protocol to those calculated on O-MAR corrected images (3 %/3 mm criteria). A typical clinical acceptance for the verification of treatment plans is 95 %. It can be concluded that the dose distributions calculated on the O-MAR corrected images are clinically identical to those calculated using the currently accepted method. Four of the 11 patients included in this study had bilateral implants. Considering a gamma criteria of 2 % and 2 mm, the four patients with bilateral hip implants accounted for 4 of the 5 lowest gamma pass rates. However, the range of pass rates was <0.009, and the 1 % error associated with the Monte Carlo dose calculations may contribute to this minor discrepancy. Therefore, it is concluded that there is no significant difference in gamma pass rates between patients with bi and unilateral hip implants.
The IAEA suggest that the CT number should be within ±20 HU of the manufacturer baseline to ensure a dosimetric accuracy of <2 % [
Assessment of the variation in CT scanner performance (image quality and Hounsfield units) with scan parameters, for image optimisation in radiotherapy treatment planning.
]. As seen from the results in Fig. 3, the HU difference between the standard and corrected images is consistently under ±70 HU, regardless of tissue. This supports the finding of negligible dosimetric differences in Fig. 4. Both results indicate that the O-MAR algorithm can be implemented clinically for dose distribution calculations for pelvis treatments with no change to the currently accepted clinical outcomes. Employing this algorithm can reduce the time to plan these treatments as the artefact volumes no longer need to be identified, contoured, and checked.
In conclusion, implementing the O-MAR algorithm resulted in CT numbers within 70 HU of the baseline values for assessed materials. Dose distributions calculated on the corrected and standard images were clinically identical. Using this reconstruction algorithm significantly reduced the noise in metal artefact affected images, improved the accuracy of contouring, and reduced inter-observer variability. The results presented show that the O-MAR algorithm can be safely clinically implemented. It can streamline the patient pathway and benefit patient outcomes by reducing planning time, improving structure visibility, and increasing contouring accuracy, and has therefore been clinically implemented in the department. In addition, an individual QA process that allows O-MAR to be considered for use on non-pelvis sites (e.g. spine, limb) on a patient-by-patient basis has been introduced.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References
Kwon H.
Kim K.
Chun Y.
Wu H.G.
Carlson J.
Park J.M.
et al.
Evaluation of a commercial orthopaedic metal artefact reduction tool in radiation therapy of patients with head and neck cancer.
Metal artefact reduction for orthopedic implants (O-MAR). White Paper, Philips CT Clinical Science, Andover, Massachusetts. 2012. Available at: https://www.philips.co.uk/healthcare/education-resources/publications/hotspot/O-MAR-metal-artefact-reduction.
Reduction of artefacts caused by orthopedic hardware in the spine in spectral detector CT examinations using virtual monoenergetic image reconstructions and metal-artefact-reduction algorithms.
Koehler T, Brendel B, Brown KM. A new method for metal artefact reduction. In: The Second International Conference on Image Formation in X-ray Computed Tomography, June 24-27, 2012, Salt Lake City, Utah, USA; authors version; 2012.
CT radiation dose reduction in patients with total hip arthroplasties using model-based iterative reconstruction and orthopaedic metal artefact reduction.
Quantitative analysis of orthopedic metal artefact reduction in 64-slice computed tomography scans in large head metal-on-metal total hip replacement, a phantom study.
Assessment of the variation in CT scanner performance (image quality and Hounsfield units) with scan parameters, for image optimisation in radiotherapy treatment planning.
00097-5&id=gr1.jpg){kind=link}
00097-5&id=gr2.jpg){kind=link}
00097-5&id=gr3.jpg){kind=link}
00097-5&id=gr4.jpg){kind=link}