The Diagnostic Accuracy of Low-Dose Computed Tomography in Diagnosing Urolithiasis: A Systematic Review and Meta-Analysis

Background: Urolithiasis is a prevalent health issue all over the world,To evaluate the diagnostic accuracy of low-lose computed tomography(LDCT) for detecting urolithiasis. Methods: Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) was followed. PubMed, EMBASE and The Cochrane Library were searched for original diagnostic studies to identify all relevant studies published prior to May 2020. The index test was LDCT, and the reference standards were comprehensive diagnosis or standard-dose CT (SDCT). Results: 17 studies with 1,761 patients and 2,053 stones were included for the quantitative analysis. The pooled sensitivity was 0.95 (95%CI: 0.93-0.97) in patient-based studies and 0.86 (95%CI: 0.76-0.93) in urolithiasis-based studies. The pooled specicity of LDCT were 0.97 (95%CI: 0.95-0.99) in patient-based studies and 0.98 (95%CI: 0.63-1.00) in urolithiasis-based studies. The Fagan nomogram of LDCT for diagnosis of urolithiasis showed that the probability of urolithiasis is 98% if the LDCT scan is positive and 6% if the LDCT scan is negative. The likelihood ratio plot showed that the summary positive pooled likelihood ratio (LRP) and negative likelihood ratio (LRN) for LDCT was in the left upper quadrant(LUQ) area. Conclusions: LDCT has excellent diagnostic value in urolithiasis. LDCT can detect the urolithiasis specically, but is limited to differentiate the contents of the stones.

urolithiasis-based studies. In the patient-based studies, LDCT was performed to con rm the diagnosis of urolithiasis in suspected patients, and number of con rmed cases were reported. This is to say, LDCT was used to diagnose whether the suspected patients were urolithiasis in the patient-based studies. In the urolithiasis-based studies, LDCT was performed to con rm whether the suspected stones were true positive urinary stones, which means LDCT was used to differentiate whether the images were truly stones rather than phlebolith or others, and the number of stones were reported. Therefore, the purpose of this meta-analysis was to evaluate the sensitivity and speci city of LDCT for the diagnosis of urolithiasis from both patient-based and urolithiasis-based studies. Additionally, subgroup analyses for ultra-lowdose computed tomography(ULDCT) (average effective dose 1 mSv) and LDCT were performed in both patient-based and urolithiasis-based group.

Methods
This meta-analysis was performed and written in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses-Diagnostic Test Accuracy(PRISMA-DTA) [12], Statement.

Literature Search
Two authors (S. Liang and L. Chang) searched on PubMed, EMBASE and The Cochrane Library from January 2000 to May 2020 independently by using the terms: "urinary calculi" OR "urolithiasis" OR "ureteral calculi" OR "ureterolithiasis" OR "kidney calculi" OR "nephrolithiasis" OR "colic", "sensitivity and speci city" OR "diagnostic accuracy", "Computed tomography" OR "low-dose CT". The articles were included by the following inclusion criteria: (a) prospective comparative studies which performed LDCT to diagnose urolithiasis in comparison to acceptable reference standard; (b) the studies reported the sensitivity, speci city and True positive (TP), False positive (FP), False negative (FN), True negative (TN); (c) the studies were written in English. Studies with insu cient or unspeci c data were excluded.
Data extraction and quality assessment Data were extracted by the same two independent readers (S. Liang and L. Chang) who performed the literature search and study selection. Disagreements were solved by a third reader (X. Li), who is a specialist in urinary imaging.
The two reviewers extracted the following information independently: the rst author, the published year, the inclusion interval of patients, reference standard of urolithiasis, the characteristics of urolithiasis patients (number, average age, BMI and stone number), the characteristics of LDCT(slice, reconstruction technique, mAs, kVp and average dose) the values (true positives, false positives, true negatives, and false negatives) and urolithiasis character from each included study. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 tool was then applied by S. Liang, L. Chang, and C. Wang independently. When there were differences in QUADAS scores between S. Liang and L. Chang, X. Li served as a third reviewer to settle discrepancies. The QUADAS is a 14-item scale designed to assess the quality of studies of diagnostic accuracy included in meta-analysis. The results of the quality assessment were summarized with RevMan version 5.3.5 (The Cochrane Collaboration, Copenhagen, Denmark)

Meta-Analysis
The index test was LDCT, and the reference standards were comprehensive diagnosis (including clinical passage, surgical removal, other imaging, other clinical information) or standard-dose CT (SDCT). The studies were divided into patient-based and urolithiasis-based group. Subgroup analysis of LDCT(average effective dose range 1 mSv to 3.5 mSv) and ULDCT(average effective dose 1 mSv) was also divided into patient-based and urolithiasis-based group. The sensitivity, speci city, positive likelihood ratio (PLR), negative likelihood radio (NLR), diagnostic odds ratio (DOR) and 95% con dence intervals (95%CI) for the accuracy of LDCT were pooled via Stata version 14.0 (The StataCorp LP, Texas City, USA). The high pooled DOR relfects the high accuracy.
We used Deek test to examine the publication bias of the included studies. Stata version 14.0 (The StataCorp LP, Texas City, USA) were used to draw the forest plots providing the assessments of sensitivity and speci city, and summary receiver operating characteristic curves(SROC) in order to acquire the area under the curve (AUC) re ecting the diagnostic accuracy of LDCT. Results were compiled using PRISMA.

Quality Assessment
All of the 17 included articles were prospective studies. Generally, most of the quality indicators of the QUADAS-2 tool were met in these included studies.
The graphical display of the risk of bias and applicability assessment is shown in Supplemental Fig. 2, the details are shown in Supplemental Table 1. 5 studies [15,20,23,29,32] were at unclear risk of bias in patient selection, of which, 4 studies [15,23,29,32] were due to the fact that the diagnosis of some patients were unclear and whether the patient samples enrolled met the criteria for continuance was not reported. Besides, one study [20] differentiated the patient selection inappropriately since patients already underwent CT scan of known urinary stones. What is more, 12 studies [15-21, 23, 25, 29-30]were at unclear risk of bias of concerning the results of LDCT examination because that the use of threshold was unclear. 4 studies [16,17,20,28] were at unclear risk of bias in the ow and timing for the reasons that the information about the duration interval was unclear. There was unclear applicability concerns of patient selection because inappropriate exclusions were avoided in one study [22]. The conduct or interpretation of the index test had unclear risk in 3 studies [15,23,29] because whether the results of the index test obtained with the knowledge of the reference standard is unclear. There was unclear applicability concerns regarding the reference in the 2 studies [23,29], because whether the results of the reference standard obtained with the knowledge of the index test is unclear.because the results comparing without the knowledge of the results of the index test. The result of Deek test to examine the publication bias of the included studies is that p = 0.98 > 0.1, which represents that there is no public bias (Supplemental

Data Extraction and Meta-Analysis
Study characters were shown in Tables 1 and 2 × 2 table, sensitivity, speci city and prevalence was shown in Supplemental Table 2. Taking all the 17 studies into consideration, the sensitivity estimates ranged from 0.70 to 0.98 and the speci city estimates ranged from 0.39 to 1.00. The result of heterogeneity test showed that 3 studies [20,26,30] have obvious heterogeneity (Supplemental Fig. 4).   Table 3.
The pooled sensitivity of the patient-based group was 0.97 in ULDCT and 0.94 in LDCT, respectively. The pooled speci city of the patient-based group was 0.95 in ULDCT and 0.97 in LDCT, respectively. The pooled sensitivity of the urolithiasis-based group was 0.85 in ULDCT and 0.90 in LDCT, respectively. The pooled speci city of the urolithiasis-based group was 0.85 in ULDCT and 0.96 in LDCT, respectively. The numbers were shown in Supplemental Table 4.
The Fagan nomogram of LDCT for diagnosis of gout showed the PLR and NLR, presenting the result that the probability of urolithiasis is 98% if the LDCT scan is positive and 6% if the LDCT scan is negative (Fig. 3), which means providing a patient is from a population with an average of 50% pre-test probability of urolithiasis, then a positive test result yields on average a post-test probability of 98%, and a negative test result yields on average a post-test probability of 6%. The likelihood ratio plot showed that the summary pooled likelihood ratio positive(LRP) and likelihood ratio negative(LRN) for LDCT was in the left upper quadrant(LUQ) area, which means LDCT can both exclude and con rm the urolithiasis (Fig. 4).

Discussion
The pooled sensitivity and speci city re ect that LDCT could be a useful test for diagnosing urolithiasis, with no major difference in the reference standard. What is more, ULDCT also has an excellent clinical use according to the subgroup analysis. However, we found that the speci city appears to be low in line with the average effective dose. In our study, ULDCT was de ned as effective dose < 1.0 mSv. Similar to ours, a systematic review de ning ULDCT as effective dose < 1.9 mSv reported that LDCT and ULDCT have high diagnostic accuracy, sensitivity and speci city despite signi cant radiation dose reduction in comparison to SDCT [31], though they may not be as effective in detecting stones < 3 mm in size or in patients with a BMI of > 30 kg/m 2 [32]. Both SDCT and LDCT have high diagnostic accuracy of ureteral uric acid stones, while the detection of uric acid stones is reduced when LDCT is at ≤ 15mAs [33]. Nevertheless, how low can the effective dose actually be still remains unknown.
In addition, the subgroup analysis of patient-based and urolithiasis-based studies shows that the pooled sensitivity of urolithiasis-based studies is much lower than that of the patient-based studies, indicating that the LDCT diagnoses urolithiasis accurately. However, LDCT may be problematic with small stones and evidence of distal ureteral obstruction [34]. The size and location of the stones may cause falsepositive and false-negative ndings, affecting the sensitivity and speci city. In the studies reporting the size of stone, the sensitivity declines when the size of stone declines [14,17,19]. Nevertheless, the thresholds of the stone were 5mm [14], 3mm [19] and 2mm [17] respectively. The speci c value of the stone threshold remains uncertian. Phlebolith may cause the false-positive and false-negative ndings, especially in distal ureter and VUJ [20,28]. What is more, heterogeneity of renal stroma may also make it di cult to differentiate between hyperdense pyramids and small calculus, thus causing false-positive ndings [19]. In additionm, passed off calculus should be taken into consideration as no calculus was detected on CT with haematuria and positive urine analysis [28].
The quality of the CT images may signi cantly degrade due to the lower radiation dose, therefore reconstruction techniques were implied in CT images. LDCT of urolithiasis can be feasible in overweight patients with a BMI between 25 and 35 kg/m 2 with iterative image reconstruction algorithms [35]. Time has witnessed the development of the reconstruction algorithms during these years. Conventional ltered back projection (FBP) reconstruction technique has been well received due to its quick imaging technology for decades [36]. Iterative reconstruction (IR) algorithms have been introduced to improve image quality with less image noise [37][38]. Advanced modeled IR (ADMIRE), adaptive statistical IR (ASIR), iterative model reconstruction (IMR), sinogram-a rmed iterative reconstruction (SAFIRE), are widely used in current clinical eld [39]. Additionally, arti cial intelligence (AI), which uses various data mining algorithms such as machine learning, deep learning, and cascading convolutional neural network model, has also been widely applied as feasible and highly-accurate ways in the diagnosis of urolithiasis [40][41].
The rst step of managing urolithiasis is accurate diagnosis, and after it is to gure out the components of the stones [42]. The most common urinary stones, calcium oxalate stones, accounts for 75-80%, and then it is uric acid stone and cystine stone taking up for 7-10% and 1%, respectively [43]. However, in a recent meta-analysis including 21 studies found that dual-energy CT (DECT), with pooled sensitivity of 0.88 and speci city of 0.98, is an accurate test for diagnosis of uric acid stones [44]. In general, LDCT is highly accurate for diagnosis of urolithiasis, whereas DECT can accurately differentiate uric acid from non-uric acid stones.

Conclusions
LDCT has excellent diagnostic value in urolithiasis given the high sensitivity and speci city of patientbased and urolithiasis-based groups. Our meta-analysis highlights the difference of the results between patient-based and urolithiasis-based groups, indicating that LDCT can detect the urolithiasis speci cally, but is limited to differentiate the contents of the stones. At the meantime, the exact dose of ULDCT remains controversial, inadequate studies about ULDCT may contribute to inevitable error in the subgroup analysis. Therefore, future studies should focus on the composition of the stones and ULDCT.        Supplementary Files This is a list of supplementary les associated with this preprint. Click to download.