Institutional review board approval was obtained, and informed consent was waived for this HIPAA-compliant retrospective study. The radiology information system was searched for adults who had undergone unenhanced CT for urinary tract stone evaluation in the non–emergency department setting between December 2011 and June 2012. Fifty-two consecutive patients, 29 men and 23 women, with an average age of 49.4 years (range, 19–91 years), were included. During this period, the clinical stone CT protocol included a split-dose calculation, wherein technologists calculated a weight-based projected volume CT dose index (CTDI_vol) and dose-length product (DLP) based on the existing protocol. Note that the existing protocol already employed contemporary dose reduction techniques including 40% adaptive statistical iterative reconstruction (ASIR). Technologists then divided the projected values into two separate CT acquisitions at 80% and 20% dose levels. These split-dose studies were performed during the first 6 months of MBIR utilization in CT, as part of initial protocol optimization to limit radiation dose exposure in patients with known or suspected stone disease. The radiation dose of the split-dose protocol was calculated to be equal to the dose in the existing (pre-split) protocol. All studies were performed on a 64-MDCT scanner (HD750 or VCT, GE Healthcare). No oral or IV contrast material was administered. Scan range was from 2 cm above the kidneys to the symphysis pubis. Pitch was 1.375:1 with table speed of 55 mm per rotation and gantry rotation time of 0.8 seconds. Full-dose tube current was 100 mA for patients less than 160 lb, and tube current was calculated as patient weight times 0.7 for patients weighing 160 lb or more; however, tube current could be as low as 10 mA to achieve dose reduction. Tube current varied as already mentioned with fixed tube potential at 120 kVp. Scans were acquired at 1.25-mm thickness and reconstructed at 2.5-mm intervals.

CT data were reconstructed with ASIR for clinical purposes, but only the MBIR series (scans at 80% and 20% dose levels) were used for the present study. Axial series at 1.25 mm and coronal series at 2 mm from 80% and 20% dose datasets were anonymized and renamed using Medical Image Resource Center (MIRC) software (Radiological Society of North America). Deidentified series were sent to a single research workstation (Vitrea, version 6.4, Vital Images) for review.

Five abdominal fellowship–trained readers with 3–33 years of experience after residency participated in the study. Readers were blinded to dose level, patient data, and clinical outcomes. Readers were given two sets of CT studies, each containing one examination for each patient with a combination of 80% and 20% dose examinations, to be rated at two reading sessions separated by at least 2 weeks. Readers had access to the axial and coronal series for each study and were asked to count the number of stones in three size categories (< 3 mm, 3–5 mm, and > 5 mm). Stone size measurements were made by each reader in preset bone windows (width, 1500 HU; level, 400 HU), targeting the largest measurement in either the axial or the coronal plane. Readers also rated each study for the following features: detectability of secondary signs of stone disease; and overall diagnostic confidence (both on a scale from 0, nondiagnostic, to 4, excellent). Specific descriptors for secondary signs (e.g., collecting-system dilatation, perinephric or periureteric stranding, and renal enlargement) were as follows: 0, poor image quality, not diagnostically acceptable for interpretation; 1, suboptimal image quality, worse-than-acceptable quality; 2, acceptable image quality, diagnostic interpretation possible; 3, good image quality; and 4, excellent image quality. Specific descriptors for diagnostic confidence were as follows: 0, no confidence; 1, findings poorly seen, low confidence; 2, findings adequately seen, confidence but with reservations; 3, findings well seen, good level of confidence; and 4, findings clearly seen, high confidence.

Anteroposterior and transverse patient dimensions were recorded, and quantitative noise measurements were made on axial images from scans at 80% and 20% dose levels by a single nonblinded reader. Noise was calculated by recording the SDs of CT number in the right hepatic lobe, subcutaneous fat, and left psoas muscle for each patient at both dose levels. Study data including CTDI_vol and DLP were recorded for each patient at each dose level, and size-specific dose estimates were calculated using CTDI_vol, patient diameter, and a conversion coefficient based on the American Association of Physicists in Medicine (AAPM) methods [17, 18]. Effective radiation dose in millisieverts was calculated by multiplying the DLP by a conversion coefficient, k (i.e., normalized effective dose [ED] per DLP over various body regions for adults, based on a 32-cm body phantom). Scans included both abdomen (k = 0.015) and pelvis (k = 0.019); therefore, an average value (k = 0.017) was used [19].

Patient data including age, sex, weight, and height were tabulated by an author. Reference standard for body mass index (BMI; weight in kilograms divided by the square of height in meters) was based on the Centers for Disease Control and Prevention (CDC) guidelines [20], with BMI below 18.5 considered as underweight, BMI of 18.5–24.9 as normal weight, BMI of 25–29.9 as over-weight, and BMI above 30 as obese.

Pearson correlation was calculated to assess concordance of stone detection between 80% and 20% dose levels for each reader. Cronbach's alpha value and intraclass correlations were used to assess interreader reliability. Chi-square tests were used to compare categoric variables, and t tests were used to compare means of continuous variables. The Mann-Whitney U test was used to compare the distribution of ratings for assessment of secondary signs of stone disease and diagnostic confidence in various subsets of patients. A value of p of less than 0.05 was the criterion for statistical significance. All statistical procedures were performed using Excel (Microsoft) and SPSS (version 21, SPSS).

The mean subject BMI was 27.5 (range, 17.3–48.7), with 11 (21%) patients over-weight and 18 (35%) obese by CDC criteria. Mean patient effective diameter, defined as the square root of patient anteroposterior times transverse diameters, was 33.3 cm (range, 21–44). Dose information for 80% scans and 20% scans is presented in Table 1, with significantly lower dose values for 20% dose scans for all dosage parameters. Thirty-four of 52 scans (65%) at 20% dose level had an ED of 1.0 mSv or less (mean, 0.77 mSv), whereas 18 of 52 scans (35%) had an ED greater than 1 mSv (mean, 1.32 mSv).

Objective noise measurements showed significantly greater image noise in all anatomic regions in scans at the 20% dose level compared with those at the 80% dose level, and subjective reader ratings for detectability of secondary signs of stone disease and overall diagnostic confidence were also significantly lower for the scans at the 20% dose level (Table 1). Only a single reader scored any studies as nondiagnostic—three at the 20% dose level and two at the 80% dose level. Overall diagnostic confidence and detectability of secondary signs of stone disease were strongly correlated in scans at both the 80% and the 20% dose levels (Table 2). There was a strong correlation between objective noise measurements and BMI at both dose levels, but there was no statistically significant correlation between diagnostic confidence and noise, BMI, or patient effective diameter (Table 2). The Mann-Whitney U test revealed no statistically significant difference in distribution of detectability for secondary signs or diagnostic confidence ratings between obese (BMI > 30) and nonobese patient groups or between overweight (BMI > 25) and nonoverweight patient groups.

Among all 52 patients, readers detected an average of 193.6 ± 25.0 stones at the 80% dose level and 154.4 ± 15.4 stones at the 20% dose level (p = 0.03) (Table 3). Agreement increased with stone size when the 80% and 20% dose level results were compared for each individual reader (Table 3 and Fig. 1). In comparison of all five readers, intra-class correlation coefficients (ICCs) also increased with stone size, with respective ICCs of 0.67, 0.69. 0.73, and 0.90 for 80% dose level and 0.70, 0.69, 0.76, and 0.88 for 20% dose level for stones less than 3 mm, 3–5 mm, and greater than 5 mm and all stones combined (p = 0.05 for < 3-mm vs > 5-mm stones). With the scans at 80% dose level for each reader as reference standard, mean sensitivity of scans at 20% dose level for detection of all sizes of stones was 86.7% (Table 4). There were no statistically significant correlations between reader years of experience, diagnostic confidence at either dose level, personal agreement of 80% and 20% dose readings, or total number of stones counted.

Our study showed excellent intrareader agreement for stone detection (Pearson correlation, 0.90) (Table 3) between aggressively dose-reduced renal stone CT (for scans at 20% dose level, mean ED = 0.97 mSv) and the low-dose reference standard (for scans at 80% dose level, mean ED = 3.9 mSv). Pooled sensitivity and specificity of 20% dose readings for detection of stones of 3 mm or larger were 92% and 82%, respectively (Table 4); mean image quality was rated as diagnostically acceptable; and mean diagnostic confidence was considered adequate but with reservations. Image noise was significantly greater in scans at 20% dose level and, although highly correlated with BMI, was not significantly correlated with diagnostic confidence. More than half of the patients in the study group were overweight or obese per current CDC criteria [20].

For all readers, the sensitivity for detecting stones of 3 mm or larger was greater than for any individual stone size grouping (Table 4). These apparently contradictory results arise from differences in stone measurements between scans at 20% dose level and those at the 80% dose level that were used as the reference standard—that is, some stones appeared slightly smaller in the higher-dose images and were categorized differently when readers counted stones. This “stone drift,” artificially lowers the sensitivity for individual size groups but is minimized when all stones of 3 mm or larger size are evaluated in aggregate.

Because most small (< 5 mm) stones pass spontaneously, it is likely that the majority of clinically significant stones (≥ 3 mm) would be appropriately detected by this aggressively dose-reduced technique. Stone CTs performed at an ED comparable to that of one or two abdominal radiographs thus offer the capability to detect and localize clinically significant stones, with adequate detectability of secondary signs of stone disease such as pelvocaliectasis and perinephric edema. These findings are concordant with prior reports on the accuracy of stone detection in dose-reduced CT scans [10–12]. However—and important to the generalizability of these findings in the North American population—our study does not limit the weight range of included patients, which is an issue in some CT dose research publications. Two recent publications have explored stone CT and included patients with similar body habitus, evaluating ASIR [13] and Siemens SAFIRE [16] but at higher CTDI_vol of 1.8 and 2.4 mGy, compared with 1.1 mGy in the current study.

Subjective assessments of diagnostic confidence and secondary signs of stone disease were less favorable for the scans at the 20% dose level. Noise measurements were objectively higher in the scans at the 20% dose level. However, on a per-patient basis, the subjectively reduced image quality was not significantly correlated with noise, BMI, or patient effective diameter. Similarly, Pick-hardt et al. [14] reported a lack of correlation between subjective image quality, objective noise measurements, and diagnostic performance in their MBIR trial performed at ED levels similar to those in the current study (for unenhanced studies, mean ED = 0.89 mSv) and recommended that additional image quality metrics should be developed and assessed for evaluation of aggressively dose-reduced images with iterative reconstruction. Anecdotally, MBIR users are well aware of an altered image texture distinct from noise, which is poorly defined and quantified thus far, but tends to be less marked in coronal reconstructed images (Fig. 2). Limited reports on this topic to date suggest retained diagnostic accuracy for high-contrast tasks such as stone detection [14] and CT angiography [21] in spite of this subjective decrement in image quality.

Our findings have potential clinical importance for patients with urolithiasis and are of particular relevance for young—and relatively more radiosensitive—patients with chronic stone disease. Such patients may undergo numerous repeat examinations; a mean annual ED of 29.3 mSv (range, 1.7–77.3 mSv) was reported for patients with urolithiasis at a tertiary care stone clinic in Canada [5]. At the same time, the prevalence of stone disease in North America is increasing quite rapidly, in proportion to metabolic syndrome and obesity [22]. Thus, the ability to detect most clinically significant stones in an approximately 1-mSv CT scan can facilitate dose reduction, on both an individual patient and a population basis.

Our study has several limitations, including the lack of an independent standard of reference for comparison. The scan at the 80% dose level, itself a low-dose study by current standards at 3–4 mSv, was used as the reference examination. However, there is no particular dose setting that can be considered as a reference standard, and the 80% dose level is in accordance with our current clinical practice. This study was not designed to compare MBIR with ASIR or FBP, and we are unable to comment on the relative efficacy of these reconstruction techniques for optimizing the image quality of dose-reduced scans. The study was limited to non–emergency department patients because the prolonged postprocessing time for MBIR—currently approximately 1 hour—limits the use of MBIR to applications that are not particularly time sensitive. Finally, given the lack of diagnostic certainty for calculi smaller than 3 mm (Fig. 3), this would be a poor technique to follow tiny stone fragments after lithotripsy.

In summary, MBIR reconstruction of aggressively dose-reduced stone CT (~1 mSv) enabled detection of the vast majority of clinically important stones (≥ 3 mm) in patients of all sizes, despite increased noise and diminished subjective diagnostic confidence. Given its low ED, this technique may be particularly useful for following known stones and imaging in radiation-sensitive populations, and we are in the process of implementing this as standard procedure in our department.

We gratefully acknowledge James Good, for assistance with image anonymization and storage, and Xi Xia, for statistical advice.