CT radiation dose reduction in patients with total hip arthroplasties using model-based iterative reconstruction and orthopaedic metal artefact reduction

Our study reveals that CT numbers remained constant, noise decreased and CNRs between muscle and fat increased in RD IMR reconstructions compared with FD iDose⁴ reconstructions for all levels of IMR while reducing CT radiation dose by 20%, 40%, 57% or 80%. By reducing metal artefacts, O-MAR reduced noise, improved CT number accuracy by bringing these values closer to expected CT numbers in the bladder and improved CNRs between muscle and fat. However, subjective image quality scores were lower in RD IMR level 1 with O-MAR images compared with FD iDose⁴ level 4 with O-MAR images, regardless of the degree of radiation dose reduction.

Large head metal-on-metal THAs composed of a cobalt chromium molybdenum alloy caused severe metal artefacts, mainly because of extensive photon starvation. These artefacts were reduced, which resulted in an improvement of CT number accuracy in the bladder and a reduction of noise. O-MAR reduced artefacts in the bladder by increasing CT numbers and reducing noise, as illustrated in Fig. 3 (p < 0.001). O-MAR also had a positive effect on noise measured in muscle and fat, which resulted in higher CNRs on O-MAR images compared with no O-MAR images, as CT numbers remained constant. Quantitative results with respect to metal artefact reduction are in concordance with a previous phantom study [13] and other patient and phantom studies, also focusing on the CT imaging of THAs [9, 10, 14‐17]. Despite the fact that O-MAR improves image quality by reducing metal artefacts, we recommend additionally evaluating conventional CT images without the use of MAR software, because MAR algorithms may introduce secondary artefacts and could degrade the depiction of bone trabeculae and bone cortex [11, 12, 18‐22].

Filtered back-projection is a fast and robust reconstruction technique, but it cannot deal with artefacts and noise when reducing radiation dose (Fig. 7). We addressed noise as the SD of pixel intensities within an ROI, and are aware that both noise and artefact affect the SD. However, in a previous study we showed that noise reduction by O-MAR is mainly caused by a reduction of metal artefacts resulting in a lower SD, as O-MAR has no influence on images without any metal artefacts [9]. As CT numbers remained similar and noise decreased, CNR improved in RD IMR reconstructions compared with FD iDose⁴ reconstructions with and without the use of O-MAR (Fig. 5). As described in a previous phantom study, lower noise values and higher CNR values are found on RD images compared with FD images, even on −80% RD images [13]. Boudabbous et al. and Kuya et al. found that MBIR furthermore reduces the size of artefacts, and that MBIR allows equal or better visibility of the bone–metal interface and improved assessment of the soft tissue surrounding implants compared with FBP [4, 5]. Yasaka et al. also found that streak artefacts were reduced using MBIR [6]. In contrast to these findings, no distinct metal artefact reduction was observed when using IMR (Figs. 7, 8).

Mean Likert scores (rated from 0 to 3), with respect to artefact and diagnostic quality of 1.59 and 1.56 were found for observers 1 and 2 respectively. In general, image quality was inferior in RD IMR results compared with FD iDose⁴ results. Low Kappa values of 0.16 and lower indicated poor or no agreement between the two observers in RD and FD images. Differences in subjective image quality scoring are not uncommon as there will be a variation in the perception of image quality and viewing strategy among observers, which was also found by Kataria et al. [23]. Despite the fact that there were differences in subjective image quality scores between observers, it is more important to focus on differences between RD and FD scores within observers. For observer 1, the image quality of RD IMR images was rated as inferior to FD iDose⁴ images with respect to: artefacts and diagnostic quality (p < 0.001), delineation of bony structures (p < 0.001), evaluation of bone density, destruction and sclerosis (p < 0.001) and the recto-uterine pouch, the vesico-uterine pouch and the bladder wall (p < 0.01; Fig. 6). Similar image quality scores were seen in FD and RD results regarding the evaluation of prosthetic components, pseudo-tumours and the evaluation of relevant hip musculature. Regarding observer 2, the quality of RD IMR images was rated as inferior compared with FD iDose⁴ images on all seven aspects (Fig. 6). Image quality scores were statistically different between the two observers regarding aspects 3, 5 and 7 on FD iDose⁴ images and aspects 5 and 6 on RD IMR images. In most of these aspects, image quality scores of observer 2 were lower compared with observer 1. This can be explained by the fact that observer 2 was less experienced in evaluating IMR images. Second, observer 1 is more experienced in evaluating metal-on-metal THAs and pseudo-tumours.

When focusing on the differences in image quality scores between RD IMR results and FD iDose⁴ results, agreement among observers improved. In most cases, image quality of FD and RD images was rated equal for observer 1 (64.7%) and observer 2 (61.5%). In 25.9% and 30.5% of the cases the image quality of the RD scan was rated 1 point lower compared with the FD scan for observers 1 and 2 respectively. In only 1.7% and 0.9% of the cases was the RD image quality rated 2 points lower compared with the FD scan for observers 1 and 2 respectively. Higher image quality scores were seen in 7.7% and 7.1% of the cases for observers 1 and 2 respectively.

Image quality of RD images reconstructed with IMR is rated lower than images reconstructed with iDose⁴, regardless of the amount of radiation dose reduction. Nevertheless, it can easily be observed that image quality is insufficient in RD IMR results at 80% reduced radiation dose, especially in bilateral metal-on-metal THAs (Fig. 8). The combination of IMR and O-MAR results in a blurring or smoothening effect, especially at the level of the large head of the THA. When there is no or little metal involved, for example in the stem of the prosthesis, this effect was not observed (Figs. 10, 11). When dealing with bilateral metal-on-metal THAs, the use of IMR combined with aggressive radiation dose reduction is therefore discouraged.

We acknowledge that commonly used metrics of image quality, noise and CNR have limited utility in fully assessing image quality for IR and MBIR algorithms [24‐26]. Samei et al. suggested that MBIR can potentially make better use of projection data to reduce CT dose by approximately a factor of 2 and furthermore found that MBIR shows improved spatial resolution for high-contrast tasks, but reduced performance for low-contrast tasks at an RD, which may influence low-contrast object detectability [25]. They showed that MBIR has a dose reduction potential of 46–84% compared with ASIR 50%, which is comparable with iDose⁴. MBIR techniques improve spatial resolution while reducing noise at the same time [1‐3, 24, 27‐29]. We quantitatively assessed CT image quality by measuring CT numbers, noise and CNR, but did not assess spatial resolution. Millon et al. assessed spatial resolution, low-contrast detectability, noise and CNR in a Catphan phantom and chest cadaver study using the same 256-slice iCT scanner and MBIR technique [28]. By determining edge-spread functions, the modulation transfer function, which is a measure of spatial resolution, was calculated. They found that MBIR improved the low-contrast detectability and enabled a CT radiation dose reduction. At an RD, spatial resolution became dose- and contrast-dependent. With an 80% radiation dose reduction, an increasing loss of structures on IMR images was observed and small low-contrast structures started to disappear, which was the case for both FBP and IMR [28]. Therefore, when applying large dose reductions, diagnostic performance could be compromised.

In our opinion, radiation dose can be reduced in THA imaging using MBIR combined with O-MAR, depending on the diagnostic purpose. Especially when evaluating the correct placement and status of prosthetic components in situ, RD protocols are sufficient. We have to keep in mind that image quality is already impaired when imaging a large-head MoM prosthesis; thus, it is unrealistic to expect high image quality scores anyway, especially at the height of the prosthetic head. Despite the fact that image quality was lower in RD IMR results, most soft-tissue abnormalities can be assessed, albeit with reduced diagnostic confidence. Mixed results were found regarding the use of MBIR techniques in soft tissues by other studies. Vardhanabhuti et al., Pooler et al. and Padole et al. reported that small lesions could be missed on RD images using IR and MBIR. Diagnostic confidence and reader confidence for detecting low-contrast lesions were reduced in RD MBIR results compared with the standard dose [7, 30, 31]. In high-contrast regions, a greater reduction of CT radiation dose may be applicable [30].

In a phantom study, Subhas et al. investigated lesion detectability near hardware using MBIR combined with MAR software versus FBP at reduced radiation dose levels and found that a 50% CT radiation dose reduction did not compromise the accuracy compared with FBP. They found that MBIR with MAR was also significantly more sensitive than FBP in detecting smaller lesions and lesions near large amounts of metal artefact [32]. Kataria et al. found that a radiation dose reduction can be applied; however, the diagnostic confidence in evaluating low-contrast objects and details in the liver was impaired using MBIR [23]. It is difficult to compare the outcomes of these studies with our own findings as the number of patients with abnormalities was limited in this population. With respect to evaluating the delineation of bony structures and assessment of bone density, bone destruction and sclerosis, the use of FD iDose⁴ reconstructions is advised. High-contrast details disappear, which could result in the loss of small details and structures such as fractures.

This study has limitations. Observers were blinded to iDose⁴ and IMR results, but one could easily identify IMR images owing to their plastic appearance, which was also encountered in other studies [7, 23]. An extensive training session for both observers could have enhanced the agreement between the observers. Radiologists are used to evaluating images containing noise and adaptation time is required to get used to MBIR images. Additionally, one of the observers was more familiar with evaluating MBIR images, with evaluating metal-on-metal THAs and with the assessment of pseudo-tumour formation. As this study involves follow-up CT scans of patients with metal-on-metal THAs, a relatively low incidence of soft tissue and bone abnormalities was observed. It is therefore hard to determine if clinically relevant findings would have been missed. Figures 9–11 illustrate that findings are also assessed in RD IMR results. It furthermore needs to be noted that the results of this study are generated with the specific IMR reconstruction algorithm using a 256-slice iCT scanner. Results cannot be compared 1-on-1 with MBIR techniques or scanner hardware of other vendors.

In conclusion, CT numbers remained constant, noise decreased and CNRs between muscle and fat increased in RD IMR reconstructions compared with FD iDose⁴ reconstructions for all levels of IMR while reducing CT radiation dose by 20%, 40%, 57% and 80%. Subjective image quality of RD IMR with O-MAR images decreased compared with FD iDose⁴ with O-MAR images in patients with large metal-on-metal THAs, regardless of the degree of radiation dose reduction. Care should be taken when using IMR in combination with reduced CT radiation dose in patients with THAs, especially in the evaluation of osseous structures and when focussing on small details and structures.