[18F]FDG PET/CT versus [18F]FDG PET/MRI for the diagnosis of colorectal liver metastasis: A systematic review and meta-analysis

Purpose The purpose of our meta-analysis and systematic review was to compare the diagnostic performance of [18F]FDG PET/CT and [18F]FDG PET/MRI in colorectal liver metastasis. Methods We searched PubMed, Embase, and Web of Science for eligible articles until November 2022. Studies focusing on the diagnostic value of [18F]FDG PET/CT or PET/MRI for colorectal liver metastasis were included. Using a bivariate random-effect model, the pooled sensitivity and specificity for [18F]FDG PET/CT and [18F]FDG PET/MRI were reported as estimates with 95% confidence intervals (CIs). Heterogeneity among pooled studies was assessed using the I2 statistic. The Quality Assessment of Diagnostic Performance Studies (QUADAS-2) method was used to evaluate the quality of the studies that were included. Results There were a total of 2743 publications identified in the initial search, finally, a total of 21 studies comprising 1036 patients were included. The pooled sensitivity, specificity, and AUC of [18F]FDG PET/CT in were 0.86 (95% CI: 0.76-0.92), 0.89 (95% CI: 0.83-0.94), and 0.92(95% CI: 0.90-0.94). [18F]FDG PET/MRI were 0.84 (95% CI: 0.77-0.89), 1.00 (95% CI: 0.32–1.00), and 0.89(95% CI: 0.86-0.92), respectively. Conclusion [18F]FDG PET/CT shows similar performance compared to [18F]FDG PET/MRI in detecting colorectal liver metastasis. However, pathological results were not obtained for all patients in the included studies and PET/MRI results were derived from studies with small sample sizes. There is a need for additional, larger prospective studies on this issue. Systematic review registration https://www.crd.york.ac.uk/prospero/, identifier (CRD42023390949).


Introduction
With a 50-60% prevalence, the liver is the most significant metastatic location from colorectal cancer (CRC) (1,2). Metastases are restricted to the liver in about one-third of these individuals at the time of detection, increasing curative therapy options (3,4). Complete resection is a common treatment option for liver metastases, which is beneficial to the prognosis (5). This observation highlights the importance of correct staging or restaging of CRC for a personalized therapy decision.
Several diagnostic imaging modalities, including contrastenhanced CT, MRI, and contrast-enhanced ultrasonography (US), are currently available for CRC staging or restaging. Contrastenhanced CT is still considered the standard imaging modality for CRC staging and restaging, however, recent studies have revealed that MRI offers superior sensitivity for detecting liver metastases (6) contrast-enhanced US is a technique that depends a lot on the operator, which may explain the lower reported liver metastasis detection sensitivity (7). There have been reports of improved sensitivity with intraoperative or laparoscopic ultrasound, but this method cannot be performed without invasive procedures (8). As a result, the most effective imaging approach has not yet been defined.
Positron emission tomography/computed tomography (PET/ CT) is an established modality for evaluating local recurrence or distant metastasis of colorectal cancer, which provides specific molecular and metabolic information (9)(10)(11). In terms of tumor staging, hybrid PET/CT system improves lesion localization and interpretation of colorectal cancer compared to PET or CT alone (12). Numerous studies have shown that PET/CT has unique advantages over conventional methods for colorectal liver metastasis (13)(14)(15)(16). However, during the past decade, radionuclide imaging techniques such as hybrid positron emission tomography/magnetic resonance imaging (PET/MRI) have attracted attention as they allow well detection of cancer metastasis. Head-to-head comparison by Brendle et al. showed that PET/CT had lower sensitivity but better specificity than PET/MRI, however, the diagnostic performance of PET/CT was lower than previous studies due to the high percentage of mucinous tumors and limited spatial resolution (17 (3); liver metastasis OR liver metastases. Studies that might have been relevant were also included from the reference lists.

Inclusion and exclusion criteria
Only studies that met all of the following criteria were included (1) (3); case reports, abstract, letters, review, or meta-analysis (4); true positive (TP), false positive (FP), true negative (TN), false negative (FN) data could not be extracted. After evaluating the titles and abstracts of the articles according to the inclusion and exclusion criteria, the full-text versions of the selected articles were examined to determine if they met the inclusion criteria. Disagreements among the researchers were settled through consensus.

Quality assessment and data extraction
Using the Quality Assessment of Diagnostic Performance Studies (QUADAS-2) technique, two researchers independently assessed the quality of the included studies. Each study's risk of bias and applicability were evaluated. It includes four important domains, including (1) patient selection (2); index test (3); reference standard; and (4) the flow and timing (19). For risk of bias, the question for patient selection was if consecutive patients enrolled; The question for the index test was if the index test results were interpreted without knowledge of the results of the reference standard; The question for reference standard was if the reference standard results interpreted without knowledge of the results of the index test; The question for the flow and timing was if there an appropriate interval between index tests and the reference(3 months). For application concern, the question for patient selection was if concerns that the included patients do match the review question; The question for index test was if there were concerns that the index test, its conduct, or its interpretation differ from the review question; The question for reference standard was if there concerns that the target condition as defined by the reference standard does not match the review question. The evaluation of each study was rated as high, low, or unclear in terms of risk of bias and applicability. To settle any potential disagreements, a third reviewer was engaged in. RevMan (version 5.3) was used for the analysis.
Data extraction for all included papers was conducted separately by two researchers. The data that were extracted included (1): the author, year of publication (2); study characteristics including country, study design, analysis, reference standard (3); patient characteristics including number of patients, clinical indication, mean/median age, chemotherapy before PET (4); technical characteristics including types of imaging tests, scanner modality, ligand dose, time from injection to acquisition, image analysis, TP, FP, FN, TN. Data were manually retrieved from the literature, tables, and figures when not clearly stated. When the paper lacked sufficient information, we contacted the corresponding authors via email to request further data or clarification. Two researchers addressed their disagreements through consensus.

Data synthesis and statistical analysis
The Spearman rank correlation coefficient were used to evaluate threshold effect performance and a P value < 0.05 indicates that the threshold effect may contribute to the heterogeneity. Using a bivariate random-effect model, the pooled sensitivity and specificity for [18F]FDG PET/CT and [18F]FDG PET/MRI were reported as estimates with 95% confidence intervals (CIs). The summary receiver operating characteristic curve and area under the curve (AUC) were generated by using the summary receiver operating characteristic (SROC) model. If the 95% confidence intervals of the two modalities did not overlap, it was considered that there was a statistically significant difference in performance. If the 95% confidence intervals of the AUC of the two modalities did not overlap, it was considered that there was a statistically significant difference in performance. Multilevel mixed-effects logistic regression was used to compare the summary paired sensitivity or specificity data, adding test type (PET/CT or PET/MRI) as covariate. Likelihood ratio tests were used to obtain the statistical differences between the sensitivities and specificities of the two tests type by fitting alternative models, adding or removing the covariate term from the model (20).
Heterogeneity among pooled studies was assessed using the I 2 statistic. Homogeneity among the studies was considered to be low, moderate, or high when the I 2 value was 25%, 50%, or 75%. For PET/CT, the meta-regression analysis and leave-one-out sensitivity analysis were conducted in the case of substantial heterogeneity (I 2 ≥ 50%) to explore possible sources of heterogeneity. For PET/MRI, we did not conduct meta-regression due to the small number of included studies (less than 10), but we perform leave-one-out sensitivity analysis to find the source of heterogeneity. Deeks' funnel plot tests were used to evaluate publication bias. All analyses were conducted with Stata 15.1 and Meta-DiSc 1.4. Statistical significance was defined as a P value less than 0.05.

Literature search and study selection
There were a total of 2743 publications identified in the initial search, 1974 studies were identified after excluding 769 duplicated studies. Based on the title or abstract,1931 studies were excluded. In the remaining results, 9 articles were irrelevant, 2 used different radiotracers, 6 data not available, and 9 PET without CT or MRI. Finally, a total of 21 articles evaluating the diagnostic performance for colorectal liver metastasis including 16 articles for PET/CT (9, 10, 13-17, 21-29), and 5 articles for PET/MRI were included (17,(29)(30)(31)(32). The PRISMA flow diagram of the study selection process is shown in Figure 1. A full list of all reviewed full-text articles was shown in Supplementary Table 2.

Study description and quality assessment
The study and patient characteristics from the 21 studies covering 1036 patients were listed in Table 1 and Table 2. Technical aspects were displayed in Table 3 and Table 4. What' s more, an assessment on the quality of involved studies was carried out depending on the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. The quality assessment graph revealed high-risk bias concerns mainly concentrated on the field of patient selection (Figure 2), caused by that most of these studies did not include consecutive patients. Overall, the risk bias of the articles was considered satisfactory. The PRISMA flow chart of study selection process.     In addition, we also performed a subgroup forest plot for PET/CT in patient-based analysis (Figure 6), the results of pooled sensitivity of [18F] FDG PET/CT in patient-based analysis for colorectal liver metastasis were 0.92 (95% CI, 0.82-0.97) and specificity were 0.89 (95% CI, 0.52-1.00).   (1): When the metabolism of 18F-FDG of the lesion is higher than that of the surrounding liver tissue, the lesion is considered positive regardless of the density on CT (2); When the metabolism of the lesion is not significantly different from that of the surrounding liver tissue, the lesion with a slightly lower density on CT and not in accordance with the typical characteristics of cysts and hemangiomas is considered positive (3); When the metabolism of the lesion is lower than that of the surrounding liver tissue, the lesion is considered negative regardless of the density on CT (3) When the metabolism of the lesion is lower than the surrounding liver tissue, the lesion is considered negative regardless of the density on CT.  Dixon-VIBE; T2WI Visual: Each reader was asked to record the size and location of the lesions and to assess the probability of colorectal cancer liver metastases for each lesion by using the following five-point confidence scale: 1,definitely not colorectal cancer liver metastases; 2, probably not colorectal cancer liver metastases; 3, inde terminate; 4, probably colorectal cancer liver metastases; and 5, definitely colorectal cancer liver metastases. Prior to image interpretation, readers were instructed that scoring a lesion as 4 or 5 might be considered a positive diagnosis for colorectal cancer liver metastases. analysis showed that image analysis(P<0.001 for sensitivity, P=0.02 for specificity), study design (P<0.001 for specificity), clinical indication (P=0.02 for sensitivity, P=0.01 for specificity), and iodinated contrast medium (P<0.001 for specificity) were the possible cause of heterogeneity (Table 5 and Figure 7). For [18F] FDG PET/MRI, sensitivity analysis by excluding data from Brendle et al. demonstrated a combined specificity of 1.00(95% CI: 1.00-1.00), with no heterogeneity (I 2 = 0%) ( Table 6).

Publication bias
Deeks' funnel plot asymmetry test showed that there was a significant publication bias for [18F]FDG PET/CT (P=0.01), and no significant publication bias was observed for [18F]FDG PET/MRI (P=0.76) (Figure 8). The staging or restaging of colorectal cancer's local and distant metastases is crucial in assessing a patient's survival and risk of recurrence, which was valuable for choosing the appropriate treatment strategy. Previously, three meta-analyses on PET or PET/ CT for colorectal liver metastasis have been published. Niekel et al. conducted a meta-analysis that included only prospective trials (34). The authors reported that PET had a higher sensitivity on a perpatient basis (0.94), in comparison to both MRI (0.88) and CT (0.75). According to another meta-analysis conducted by Floriani et al., PET showed the highest sensitivity (0.94), followed by MRI (0.81), CT (0.75), and ultrasound for the diagnosis of liver metastases (6). A meta-analysis conducted by Maffione et al. indicated that PET is less sensitive (0.93) but more specific (0.93) than MRI and has an impact on the therapy of roughly one-fourth of patients (35). However, all of the previous meta-analyses didn't mention hybrid PET/MRI system, which was an effective modality in detecting colorectal liver metastasis proven by recent studies. Thus, there was an urgent need to assess the diagnostic performance of PET/CT versus PET/MRI.

Discussion
Our pooled sensitivity for PET/CT was inferior compared with the findings of both previously published meta-analyses (0.86 vs. 0.94 and 0.94). The specificity of PET/CT in our results were also inferior compared with the results of previous meta-analysis (0.89 vs. 0.94 and 0.93). This was due to our inclusion of studies that analyzed data from both patients and lesions. When we pooled sensitivity with only patient-based studies, the sensitivity (0.93) and specificity (0.90) were in line with the previous meta-analyses.
Even while [18F]FDG PET/CT and PET/MRI were increasingly being utilized to detect distant metastases in suspected cases of colorectal cancer, they were not the first-choice modality in the routine evaluation of CRC. However, Ruers et al. demonstrated the addition of PET over conventional imaging modalities, which   Graph of risk of bias and applicability of all eligible studies based on QUADAS-2 tool.

FIGURE 3
Forest plots of the combined [18F]FDG PET/CT sensitivity and specificity for colorectal liver metastasis. Squares denoted the sensitivity and specificity in each study, while horizontal bars indicated the 95% confidence interval.

FIGURE 4
Forest plots of the combined [18F]FDG PET/MRI sensitivity and specificity for colorectal liver metastasis. Squares denoted the sensitivity and specificity in each study, while horizontal bars indicated the 95% confidence interval.  Forest plots of the combined [18F]FDG PET/CT sensitivity and specificity in patient-based analysis for colorectal liver metastasis. Squares denoted the sensitivity and specificity in each study, while horizontal bars indicated the 95% confidence interval.   Multiple univariable meta-regression forest plot of [18F]FDG PET/CT for colorectal liver metastasis.
the high cost and diagnostic performance of PET/MRI, it was determined that the diagnostic value offered by PET/CT was effective enough to serve as a suitable alternative. In terms of the diagnostic performance, our findings may be applicable and timely for clinical decision-making. However, due to the small sample size of hybrid PET/MRI, further larger prospective studies focusing in the diagnostic performance of PET/MRI in colorectal liver metastasis are still needed to obtain a more robust result. In terms of heterogeneity, there was high heterogeneity in [18F] FDG PET/CT (sensitivity and specificity) and [18F]FDG PET/MRI (specificity). We examined the sources of heterogeneity among the studies by performing meta-regression and sensitivity analysis. For [18F]FDG PET/CT, meta-regression analysis showed that image analysis, study design,clinical indication, and iodinated contrast medium were the possible cause of heterogeneity. For [18F]FDG PET/MRI, we got an acceptable heterogeneity (I 2 = 0%) by excluding data from Brendle et al., which could be explained by different previous treatment and cut-off thresholds. Nevertheless, there may be further causes, such as differences in patients, technique, and study design. It yielded a same specificity (1.00) when we omitting the study by Brendle et al., which further proved the robustness of the results.
The limitations of our meta-analysis should also be mentioned. First, only five studies offered sufficient information that evaluated the diagnostic performance of the hybrid PET/MRI system for the staging or restaging of liver metastases in CRC patients, which leads to small sample size. This was because hybrid PET/MRI were introduced in the recent years and still lack of well-designed trials. Second, Due to the limited number of research that satisfied the inclusion criteria, we included either patient-based or lesion-based analysis studies, which could increase potential bias. Third, most of the reference standard for diagnosis colorectal liver metastasis was pathology and follow-up imaging, pathological results are not obtained for all patients in the included studies. These results therefore need to be interpreted with caution.

Conclusion
[18F]FDG PET/CT shows similar performance compared to [18F]FDG PET/MRI in detecting colorectal liver metastasis. However, pathological results were not obtained for all patients in the included studies and PET/MRI results were derived from studies with small sample sizes. There is a need for additional, larger prospective studies on this issue.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.

Author contributions
ZM and XZ conceived and designed the study, which were proofed by ZM. XZ and XL collected and analyzed the data. ZM wrote the manuscript. All authors contributed to the article and approved the submitted version.