Accuracy comparison of various quantitative [^99mTc]Tc-DPD SPECT/CT reconstruction techniques in patients with symptomatic hip and knee joint prostheses

The prevalence of total hip and knee arthroplasties (THA, TKA) has been constantly increasing over the last decades [1]. As a result, the number of dysfunctional or loose prostheses needing revision surgery is increasing. Reported rates of prosthetic loosening for THA and TKA range from 2% after 5 years to 8% after 15 years and of 2.3% after 5 years to 7.1% after 15 years, respectively [2]. Aseptic loosening is the most common reason for revision operations of THA (40%) and TKA (36%) [3].

In clinical routine diagnostic work up of symptomatic joint prosthesis is based on a combination of clinical examination and imaging studies. Plain radiographs are the standard imaging method for follow-up of joint replacements [4]. They are used to assess the integration, alignment and integrity of implants. However, the interrater agreement of radiographic assessment of prosthetic loosening is low [5] as is the sensitivity for the detection of periprosthetic osteolysis [6]. Recent advances of metal artefact reduction in MR imaging allow the visualization and assessment of the bone-implant interface when implants with low susceptibility for metal artefacts such as titanium alloy-based implants are used. However, MR findings such as bone marrow oedema or periosteal reaction appear also but less frequent in asymptomatic prosthetic joints limiting the specificity of this imaging modality [7].

A functional imaging modality is planar bone scintigraphy or 3D-SPECT imaging of bone remodelling that uses technetium-labelled diphosphonates, in our study ^99mTc-Dicarboxypropandiphosphate (DPD), to target immature bone matrix (osteoid). An increased tracer uptake in bone indicates higher osteoid content of bone matrix and is therefore indicative for an increased osteoblastic activity, for example, due to mechanical stress of a loose prostheses. This technique is used as a second line diagnostic imaging method for the evaluation of painful hip and knee joint prosthesis since the 1970s with diverging reports of diagnostic accuracy [8‐12]. Hybrid-scanners combining SPECT and computer tomography (CT) are used for more than 15 years. With its ability to directly correlate functional information with structural changes in morphological imaging data, this new technique is more and more replacing conventional planar scintigraphy, also for this particular diagnostic question of aseptic prosthetic loosening [13, 14]. The main focus of the diagnostic criteria of prosthetic loosening in SPECT/CT lies in the distribution of elevated bone metabolism around the prosthetic implant. Only few studies investigated the diagnostic value of uptake intensity of periprosthetic bone [15‐17] with promising results although only relative semiquantitative uptake measures were used. Recently, a new generation of SPECT/CT scanners and reconstruction algorithms of SPECT data was introduced that allows the absolute quantification of tracer uptake [18] in clinical routine. Similar to positron emission tomography (PET), quantitative uptake measures can be normalized and expressed in standardized uptake values (SUV) which potentially allows the discrimination between stable and loose prostheses. The value of absolute uptake quantification of bone SPECT/CT was already investigated for several applications in musculoskeletal and oncological imaging [19‐22]. To the best of our knowledge, no published data exist analysing the diagnostic value of truly quantitative measures of bone metabolism around prosthetic joint implants for the detection of aseptic loosening.

Therefore, we compared the diagnostic accuracies of different quantitative SPECT reconstruction algorithms (xSPECT Quant and xSPECT Bone) obtained with a Siemens Symbia Intevo T16 System [23, 24] with and without iterative metal artefact reduction (iMAR) in order to detect aseptic implant loosening of patients with symptomatic hip and knee joint prosthesis. Standard of comparison was surgical diagnosis or follow-up observation over a period of at least 1 year.

In total, 39 patients were screened. However, 9 patients had to be excluded due to periprosthetic fractures (n = 2), joint infection (i.e. microbiologically proven periprosthetic infection within 1 month after SPECT/CT, n = 1), implantation of arthroplasty < 6 months prior to the SPECT/CT referral (n = 2), lost to follow-up during the study (n = 2), incomplete follow-up data n = 1), and withdrawal of consent (n = 1) (Fig. 3). The remaining 30 patients consisted of 17 females and 13 males with a mean age of 71 years. Mean age of prosthesis was 11.6 years. The 30 patients had altogether 44 arthroplasties (Table 1). Fourteen patients had bilateral arthroplasties (7 THA, 7 TKA), and three patients had bilateral suspicion of loosening (3 TKA).

Thirty-three of the 44 prosthesis were symptomatic and suspicious for being loose. Of the 33 symptomatic arthroplasties, loosening was confirmed for 10 arthroplasties. Of these, 3 arthroplasties were diagnosed intraoperatively. The surgical diagnoses for confirmed loose arthroplasties were: “aseptic loosening of patellofemoral compound” (1 TKA), “aseptic loosening of femoral shaft” (1 THA) and “aseptic loosening of acetabular cup” (1 THA). Loosening of 7 arthroplasties was diagnosed by follow-up.

Loosening was ruled out for 23 arthroplasties. Of these, 9 were confirmed intraoperatively by finding other diagnosis than loosening. For 14 arthroplasties, loosening was ruled out by follow-up.

The remaining 11 arthroplasties were asymptomatic prostheses in patients with bilateral arthroplasty and were included in the analysis of SUVmax measurements as normal control arthroplasties.

Twenty-three arthroplasties were cemented (13 THA, 8 TKA, 2 UKA), of which 6 were loosened prostheses (3 THA, 3 TKA).

The main findings of this study are: (1) Quantitative SPECT/CT revealed a significantly increased [^99mTc]Tc-DPD uptake in loosened hip and knee joint replacements compared to stable joint replacements independent on the type of prosthesis. (2) Quantification with xSPECT Bone plus iMAR showed the highest accuracy of all evaluated methods in the discrimination between loose and stable prostheses. (3) Quantitative SPECT/CT seemed to differentiate better with a higher accuracy between loose and stable prostheses than conventional visual reading.

Only few studies investigated the relative intensity of periprosthetic bone metabolism with bone scintigraphy and bone SPECT/CT referencing the periprosthetic activity with ROIs of the femur [15‐17] and found a good interreader agreement for such a standardized protocol [16, 17]. To our knowledge, there is only one available study applying quantification of periprosthetic uptake for the diagnosis of aseptic loosening published 2008 by Klett et al. [15]. In that retrospective study, planar scintigraphies of 31 cemented knee joint prosthesis were used placing ROIs in the periprosthetic tibia and a normalizing control ROI in the femur. With this approach, the differentiation between loose and stable prostheses was possible with an accuracy of 94% which corresponds well to our results. However, only patients with revision surgery were included and it is likely that the results of the prior bone scan contributed to the decision for surgery, thus introducing a selection bias. In contrast, we used a prospective approach, including patients prior therapeutical considerations and followed them for at least 1 year if no surgery was performed. This prospective design makes the results of our study more meaningful for clinical application. Moreover, the direct quantification of [^99mTc]Tc-DPD uptake after standardized calibration of the scanner with only one VOI covering the whole prosthesis is highly feasible and may allow automated absolute quantification. The correlation with CT-morphology allows the exclusion of potentially false positive bone uptake which is not associated with the implant interface such as osteophytes, heterotopic calcifications and others. This was necessary in 25% of cases.

The comparison of diagnostic accuracies of the different quantification methods in our study showed that the combination of the xSPECT Bone with CT including iterative metal artefact reduction for attenuation correction performed best in diagnosing prosthetic loosening with an accuracy of 93.9%. Most likely, this can be explained with the higher resolution of the xSPECT Bone data compared to xSPECT Quant data reducing the partial volume effect, thus leading to a higher discrimination of different uptake levels.

Moreover, CT-metal artefact reduction with iMAR does not only improve the image quality of CT images for morphologic assessment, it also showed to improve attenuation correction around metal implants as shown in a recently published study [26]. CT-streak artefacts around the metal implant can influence the zone mapping of xSPECT bone and may lead to a false attribution of a tissue class to a voxel, as streaks artificially change the depicted density. A soft tissue may then be classified as “cortical bone” class and then obtain initially bone activity normalization. As iteration continues the activity estimate in that voxel is being reduced towards the correct soft tissue uptake, but iteration typically terminates well before then. What remains is an uptake which is higher than soft tissue, yet lower than actual bone tissue. These “shining artefacts” as described in a recent case series by Lima et al. [27] were not evident in our study; a reason may be the marked reduction of these artefacts with iMAR, which may lead to a more precise quantification and may have contributed to the high accuracy of the combination xSPECT bone with iMAR reconstructed CT in our study (Fig. 2 versus Fig. 5).

Visual reading of triple phase scintigraphies together with SPECT/CT showed a lower accuracy compared to quantitative uptake measures. Interestingly early phases did not contribute to the diagnostic decision with an accuracy of only 48% for early phases alone. The difference of accuracies of visual reading and quantitative uptake measures was, however, not significant. Nevertheless, quantification of periprosthetic uptake as proposed could add confidence to the diagnostic decision, especially for physicians with limited experience.

This study has several limitations: Due to the selection criteria, the study cohort may not be representative for the broader and more heterogeneous collective of patients with symptomatic prosthetic joints as seen in general practices and ambulant facilities. In particular, the pre-test probability may be lower compared to our cohort. As a consequence, the accuracy of quantitative SPECT/CT may be lower in this context. The small cohort size of 30 patients did not allow subgroup analysis such as knee versus hip joint prostheses or uncemented versus cemented prostheses. Still, we found highly significant differences with clearly higher uptake levels in loosened arthroplasties even in this small cohort.

We compared quantitative uptake values of four different reconstruction methods and compared several diagnostic parameters (accuracy, sensitivity, specificity, positive and negative predictive values) of these four quantification methods and the visual readings. This leads to the problem of multiple comparisons and the possibility that some of the significant differences could be produced by chance should be taken into account when interpreting our results. However, since the differences of SUVmax between loosened and stable prosthesis are very clear and highly significant and the diagnostic parameters, which are closely related to each other show a consistent tendency, we believe that the results are robust.

We included some patients (n = 4) with a postoperative time interval between 6 and 12 months. In this relatively early postoperative phase, in which implant failures can occur, a higher bone metabolism can be expected also in asymptomatic patients [28, 29]. A sensitivity analysis on the time interval between implantation and SPECT/CT did not reveal a significant influence on the results. Still in this early implantation phase periprosthetic uptake has to be interpreted with caution and SUV values may not be reliable for the diagnosis of aseptic loosening.

Also, the heterogeneity of implant models with cemented and uncemented arthroplasties might have biased the results, e.g. it was shown by Ullmark et al. that uncemented prostheses have higher and longer lasting ¹⁸F-fluoride uptake levels compared to cemented stems [30]. Again, we controlled for this bias performing a sensitivity analysis and did not find a significant influence on the results.

We did not analyse the distribution pattern of tracer uptake around implants as did other studies [16] since we wanted to evaluate [^99mTc]Tc-DPD uptake values as a quantitative biomarker for the diagnosis of prosthetic loosening independent of uptake patterns.

All evaluated quantification methods showed a significantly higher [^99mTc]Tc-DPD uptake around loose arthroplasties than around stable arthroplasties. Importantly, quantification with xSPECT Bone plus iMAR discriminates best between loose and stable prostheses. With cut-off values derived by ROC analysis, an accuracy of nearly 94% could be reached solely with quantitative uptake measures proposing [^99mTc]Tc-DPD uptake quantification as a promising biomarker for the diagnosis of prosthetic loosening. However, further studies will be necessary which will analyse and quantify periprosthetic uptake pattern in a larger number of patients with symptomatic knee and hip prostheses in order to fully utilize the potential of uptake quantification as a biomarker.