Meta-Analysis of the Correlation between Apparent Diffusion Coefficient and Standardized Uptake Value in Malignant Disease

The objective of this meta-analysis is to explore the correlation between the apparent diffusion coefficient (ADC) on diffusion-weighted MR and the standard uptake value (SUV) of 18F-FDG on PET/CT in patients with cancer. Databases such as PubMed (MEDLINE included), EMBASE, and Cochrane Database of Systematic Review were searched for relevant original articles that explored the correlation between SUV and ADC in English. After applying Fisher's r-to-z transformation, correlation coefficient (r) values were extracted from each study and 95% confidence intervals (CIs) were calculated. Sensitivity and subgroup analyses based on tumor type were performed to investigate the potential heterogeneity. Forty-nine studies were eligible for the meta-analysis, comprising 1927 patients. Pooled r for all studies was −0.35 (95% CI: −0.42–0.28) and exhibited a notable heterogeneity (I2 = 78.4%; P < 0.01). In terms of the cancer type subgroup analysis, combined correlation coefficients of ADC/SUV range from −0.12 (lymphoma, n = 5) to −0.59 (pancreatic cancer, n = 2). We concluded that there is an average negative correlation between ADC and SUV in patients with cancer. Higher correlations were found in the brain tumor, cervix carcinoma, and pancreas cancer. However, a larger, prospective study is warranted to validate these findings in different cancer types.


Introduction
At present, various imaging modalities play an important role in diagnosis, staging, follow-up, and therapeutic evaluation of patients with cancer. Positron emission tomography/computed tomography with F-18 based fluorodeoxyglucose ( 18 F-FDG PET/CT) is considered as an accurate method for characterizing tumor lesions due to the availability of anatomic and glucose metabolic information of tumor [1]. The standardized uptake value (SUV) is the most frequently used parameter derived from 18 F-FDG PET, which has been used for assessing tumor aggressiveness, differentiating benign from malignant tumors, and monitoring treatment [2,3].
Magnetic resonance imaging (MRI) is another important tool to detect and characterize tumors. Specifically, diffusionweighted imaging (DWI) provides an additional promising dimension to the conventional anatomical MRI. Apparent diffusion coefficient (ADC) is a parameter obtained by MR-DWI, reflecting the Brownian movement of water molecules. The ADC value has been shown to link with the cell density, microvascular circulation, and membrane integrity of a tumor tissue [4].
Although glucose metabolism and cell density represent two different facets of tumor biology, many researchers tried to find the relationship between ADC and SUV. However, there is a controversy in this relationship. Some data demonstrated that there was no significant correlation observed between SUV and ADC [5], while other studies reported that SUV was inversely correlated with ADC [6,7]. Given the conflicting evidence on this issue, we conducted this metaanalysis to explore the correlation between ADC and SUV in cancer patients.

Study Identification and Selection.
Two independent reviewers evaluated the potentially relevant articles on the basis of the inclusion and exclusion criteria. Articles were included if they met the following criteria: The exclusion criteria included the following: (a) Data or part of data presented in more than one article (in this case, the article containing the latest and/or the most complete data was chosen) (b) Animal studies, reviews, case report, letters, editorials, abstracts, comments, and in vitro studies (c) Studies including less than 10 patients or 10 lesions (d) Articles without sufficient information for calculation of correlation coefficient If there was discordance among the 2 independent researchers for one study, its eligibility was decided by the 3rd investigator. If more than one correlation coefficient value calculated according to various SUV indexes or ADC indexes was reported in the article, the lowest value was chosen.

Data Extraction
When disagreements occurred between the two reviewers, a third investigator joined to vote for a decision.

Methodology of Quality Assessment.
Two investigators (Shengming Deng and Bin Zhang) assessed the quality of the articles independently according to the QUADAS-2 [58], which consists of 2 parts of contents: "risk assessment" and "practical application." The former was assessed from 4 key domains as patient selection, index test, reference standard, and flow and timing, and the latter included 3 aspects as patient selection, index test, and reference standard.
To make sure that the QUADAS-2 tool is applicable to the present study, we designated SUV measurement as the "reference test" and ADC measurement as the "index test." In this study, we chose one month as the threshold interval between PET or PET/CT examination and DWI-MRI detection in case tumor biology will change much. A third reviewer was introduced when there were assessing differences between the two observers.   [59]. Publication bias was assessed by means of Begg's funnel plots and Begg's statistical test.

Meta-
The heterogeneity of values between studies was tested by calculating statistic and the inconsistency index (I 2 ). < 0.05 or I 2 > 50% indicated the presence of heterogeneity. In case of the existence of heterogeneity, a sensitivity analysis was performed for all studies to further investigate the study heterogeneity. In a subgroup analysis, studies were stratified according to tumor type and correlation coefficient value (SUV mean /ADC mean , SUV max /ADC min , SUV max / ADC mean , etc.).
Statistical analysis was performed using STATA 11 software package (Stata Corporation, College Station, TX, USA). < 0.05 was considered statistically significant.

Literature Search and Selection of Studies.
The original search identified 145 articles in PubMed and 759 articles in EMBASE. After removing duplicates, 789 abstracts were screened according to the evaluation criteria, and 115 in total were selected to be read in full as potentially eligible. After reading the full texts, 66 studies were excluded for the following reasons: (1) the article did not involve the evaluation of the relationship between ADC value and 18 F-FDG uptake ( = 38); (2) the number of cases or tumor sites studied was fewer than 10 ( = 13); (3) the raw data in the article failed to generate the correlation coefficient values ( = 10); (4) part of the data in the study appeared in other articles ( = 3); (5) parameters measured by two individual reviewers were presented in the article which was difficult to choose ( = 1); and (6) most of the cases studied were benign tumors ( = 1). Figure 1 describes the study selection process and results according to the PRISMA guidelines. Finally, 49 published articles were included in the present study .

Study Characteristics.
The selected studies were published between 2008 and 2015. The median number of patients per study was 32 (range: 7-131) with a total number of 1927 patients. In some studies, more than one tumor site was analyzed on several patients; therefore, a total of 2356 samples were assessed in the meta-analysis. Studies covering a range of cancer sites are summarized in Table 1.
The most studied tumor location was the lung with 10 studies. The second tumor type was head and neck cancer, for

The Results of QUADAS-2
Assessing the Quality of the Included Articles. As shown in Figure 2, the results of QUADAS-2 assessing the quality of the included articles indicated that the results of 10 studies adequately addressed all risk of bias domains. Among all the 49 studies, risk of bias was high or unclear with regard to patient selection for 7 studies, the index text for 32 studies, the reference standard for 31 studies, and flow and timing for 14 studies.
Interpretation of ADC or SUV in a blinded fashion was an item which most studies did not report. Seventeen studies clearly stated that the index test was assessed without knowledge of the results of the reference standard, while this was unclear in 32 studies. Similarly, in 18 studies, the interpretation of reference standard was clearly stated as under unknown index test, while the other 31 studies did not state the interpretation of reference standard clearly.
Acceptable delay between reference and index tests was the item which many studies did not report. Eleven studies provided no information about the time interval between the execution of MR-DWI and the 18 F-FDG PET/CT scan. In 3 studies documented, the interval was more than 4 weeks.
In addition, patients enrolled in 1 study were investigated on residual tumors after completion of therapy. In these patients, whether the relationship between 18 F-FDG uptake and ADC value differs from that in patients with pretherapeutic tumor is unclear; therefore, the risk of case selection bias in this study was considered unclear in the present analysis.

The Results of a Meta-Analysis.
The data provided by the finally chosen studies all met the standard of meta-analysis. values for 3 studies were calculated from provided 2 , and values for 2 other studies were determined from the provided scatter plot. For 3 other studies, values were calculated based on the provided raw data of corresponding ADC and SUV.
Final combined value calculated from all the included articles was −0.35 (95% CI: −0.42-−0.28), but the results of heterogeneity test indicated the presence of marked heterogeneity among studies (I 2 = 78.4%; < 0.01; Figure 3). We then conducted a sensitivity analysis by excluding each article at a time to observe its effect on the final outcome, but the results showed that no individual study contributed more greatly to the total heterogeneity. The results of Begg's test indicated no significant publication deviation among the included articles ( > 0.05; Figure 4).
As shown in Figure 5, the subgroup analysis for tumor types showed that combined for the 10 studies of lung cancer was

Discussion
In the recent years, the correlation between ADC and SUV has been increasingly studied. In the present study, we investigated the relationship between 18 F-FDG uptake and ADC value using meta-analysis methods. Our meta-analysis showed that, in cancer patients, there was an average negative correlation between the SUV and ADC. Subgroup analysis on different tumor types indicated that degrees of correlation among different tumor types varied and heterogeneity of some subgroups changed significantly. The subgroup analysis on various correlation coefficient values indicated that combined values of subgroups did not show significant changes, and there were no significant changes in heterogeneity.
In this study, we used QUADAS-2 as an evidence-based quality assessment tool. In the present analysis, the vast majority of the articles did not mention whether the test results of DWI-MRI and 18  ADC was not clearly stated. In addition, some studies did not address the inclusion criteria of patients adequately. The above problems may increase the bias of study.
DWI provides an excellent tissue contrast through detection of differences in the Brownian motion of water molecules in tissues. ADC is a parameter calculated from DWI and altered by any architectural changes in the proportion of extracellular to intracellular water molecules because the diffusion of water molecules is disturbed by intracellular organelles and macromolecules [60]. Malignant tumors usually show decreased ADC values because they are characterized by increased cellularity, larger nuclear/cytoplasmic ratio, and less extracellular space relative to normal tissues which restrict the diffusion of water molecules [61]. Currently, 18 F-FDG PET/CT has been considered as the standard of care in various cancers. 18 F-FDG uptake is correlated with the number of viable tumor cells and their metabolic activity. Glucose utilization in tumors is increased due to the Warburg effect [62]. Recently, the introduction of simultaneous PET/MRI makes it possible to combine functional and metabolic studies in malignancies in one examination. It was postulated that there is a correlation between ADC and SUV. The present study showed that the pooled correlation coefficient between SUV and ADC was −0.35, indicating an average negative correlation. A possible explanation of this result might be that although there is a certain overlap of the information provided by 18 F-FDG PET and DW-MRI, the two parameters (SUV and ADC) reflect different tumor biology. For example, except for cellularity, ADC is correlated directly with tumor necrosis because of increased presence of free water in the necrotic area [63]. However, 18 F-FDG PET demonstrates tumor necrosis as photopenic defects. In addition, although the ADC measurement is derived from DWI which is an MR sequence that is known for a high detection rate of lesions, it is not always very specific [64]. Our result suggested that 18 F-FDG PET and DWI-MRI might complement each other on the clinical diagnosis.
We conducted a subgroup analysis based on different tumor types. The meta-analysis about ADC and tumor cellularity correlation revealed no notable variation between the subgroups based on cancer type [65]. In this study, our results showed that the correlation between 18 F-FDG SUV and ADC differed between histological types. Combined correlation coefficients range from −0.12 (lymphoma, = 5) to −0.59 (pancreatic cancer, = 2). Correlation was moderate in brain, cervix, and pancreas, average in lung, head and neck, breast, and rectum, and poor in lymphoma. However, this issue needs to be further explored with more experiments.
The present study has some potential limitations. First, although the number of patients included in this study was large, they were relatively limited to a certain type of tumors. This may cause limitations in our inference based on the results of subgroup analysis on different histological types. Second, our meta-analysis was based only on published studies which provided values or raw data which can be used to calculate values. Other articles which only report positive or negative results without specific data were excluded from this analysis. In addition, this study was restricted to articles published in English, which would cause publication bias. However, the results of Begg's test showed no evidence of publication bias. We also used the random-effects model to reduce heterogeneity. Therefore, the results of the present study are reliable.
In short, although there are limitations in this study, our meta-analysis demonstrated an average negative correlation between the SUV and ADC values in patients with cancer. Sufficient data support a moderate correlation for brain, cervix, and pancreas, average correction for lung, head and neck, breast, and rectum, and poor for lymphoma. However, a prospective study with a larger population is warranted to validate these findings in different cancer types.

Disclosure
Shengming Deng and Zhifang Wu are co-first authors.

Authors' Contributions
Shengming Deng and Zhifang Wu contributed equally to this article.