Detection and Characterization of Renal Stones by Using Photon-Counting–based CT

Abstract

Purpose

To compare a research photon-counting–detector (PCD) CT scanner to a dual-source, dual-energy CT scanner for the detection and characterization of renal stones in human participants with known stones.

Materials and Methods

Thirty study participants (median age, 61 years; 10 women) underwent a clinical renal stone characterization scan by using dual-energy CT and a subsequent research PCD CT scan by using the same radiation dose (as represented by volumetric CT dose index). Two radiologists were tasked with detection of stones, which were later characterized as uric acid or non–uric acid by using a commercial dual-energy CT analysis package. Stone size and contrast-to-noise ratio were additionally calculated. McNemar odds ratios and Cohen k were calculated separately for all stones and small stones (≤3 mm).

Results

One-hundred sixty renal stones (91 stones that were ≤ 3 mm in axial length) were visually detected. Compared with 1-mm-thick routine images from dual-energy CT, the odds of detecting a stone at PCD CT were 1.29 (95% confidence interval: 0.48, 3.45) for all stones. Stone segmentation and characterization were successful at PCD CT in 70.0% (112 of 160) of stones versus 54.4% (87 of 160) at dual-energy CT, and was superior for stones 3 mm or smaller at PCD CT (45 vs 25 stones, respectively; P = .002). Stone characterization agreement between scanners for stones of all sizes was substantial (k = 0.65).

Conclusion

Photon-counting–detector CT is similar to dual-energy CT for helping to detect renal stones and is better able to help characterize small renal stones.

© RSNA, 2018

Introduction

Assessment of renal stones at dual-energy CT may noninvasively identify stone composition (1,2), but performance decreases for small stones (3–5). Photon-counting–detector (PCD) CT has potential advantages compared with dual-energy CT with conventional energy-integrating detectors (EIDs). Compared with EID CT, PCD CT images have an intrinsically higher signal-to-noise ratio (6), allow an energy-dependent weighting factor to be assigned to each energy bin to improve signal-to-noise ratio or reduce radiation dose (7), and use energy discrimination to provide information about the elemental composition of imaged materials (8–10).

Previous experience has shown numerous clinical benefits of PCD CT compared with EID CT, including similar image quality in the abdomen and improved image quality in the chest (11–15). We compared a research PCD CT scanner to a dual-source, dual-energy CT scanner for the depiction and characterization of renal stones in human participants with known stones.

Materials and Methods

This Health Insurance Portability and Accountability Act–compliant, prospective study was approved by our institutional review board. Written informed consent was obtained from each participant. The PCD CT system was provided by Siemens Healthcare to our institution as part of a grant from the National Institutes of Health. Authors from the Mayo Clinic (Rochester, Minn) retained control of all study information and statistical analysis, and all reported results.

Participants

Thirty participants underwent clinically indicated dual-energy CT for renal stone characterization between September 2015 and March 2016 followed by noncontrast agent–enhanced renal PCD CT (Fig 1). Only nonpregnant adults who had a lateral width at the level of the liver of less than 50 cm were included.

Figure 1:

Study flowchart. DSCT = dual-source CT, EID = energy-integrating detectors, GU = genitourinary, PCD = photon-counting detectors.

Data Acquisition

EID dual-energy CT.—A 128–detector row dual-source CT system (Somatom Definition Flash; Siemens Healthcare, Forchheim, Germany) acquired dual-energy EID CT data. Our clinical protocol for noncontrast-enhanced, dual-energy CT stone characterization was used to scan participants, with tube potential pair determined by the lateral width of the patient at the level of the liver dome, as follows: 80 kV for one x-ray tube and 140 kV with a tin filter in front of the other x-ray tube (80/Sn 140 kV) for widths 35 cm or less; 100/Sn 140 kV for widths of 36–50 cm. Other acquisition parameters were 0.5-second rotation time, 0.6 pitch, and 32 × 0.6 mm collimation, with use of automatic exposure control (CareDose 4D; Siemens Healthcare).

PCD CT.—The research PCD CT scanner used in this study is a modified version of a second-generation EID dual-energy CT system (11–13). Anterior-posterior and lateral CT localizer radiographs were acquired for renal centering within the PCD CT's 27.5-cm field of view. To avoid truncation artifacts, a 1-mGy data completion image was acquired (16). Unenhanced spiral scanning was performed by using the same parameters as the EID dual-energy CT examination, with the exception of the following: 140-kV tube potential, 25- and 75-keV energy thresholds, and 32 × 0.5 mm collimation with no automatic exposure control. The two energy thresholds were selected to give approximately equal numbers of photons in the two energy bins (bin 1 and bin 2) for patients with 35–40 cm lateral width (17,18). Tube current was adjusted so that the volumetric CT dose index before acquisition of the image matched that used for the EID examination. After completion of the PCD scan, the postacquisition volumetric CT dose index was recorded.

Data Reconstruction

Low- and high-energy EID CT images, along with EID mixed-kilovolt images (0.5 weighting factor) were reconstructed by using a quantitative kernel (D30, 1-mm thickness, 0.8-mm interval and 5-mm thickness, 2.5-mm interval) and a medium-sharp body kernel (B40, 1-mm thickness, 0.8-mm interval).

PCD data were reconstructed offline by using the same weighted filtered back-projection algorithm and reconstruction pipeline used in the reconstruction of the EID data (19). PCD images were reconstructed by using data corresponding to the low-energy threshold (T_L; 25–140 keV), bin 1 (25–75 keV), and bin 2 (75–140 keV) with identical thicknesses, reconstruction intervals, and kernels (18).

To blind readers to the CT scanner type, all images were reconstructed with a 27-cm circular field of view.

Visual Stone Detection

Two board-certified genitourinary radiologists (T.J.V., M.L.W.), blinded to CT system and image type, evaluated images to detect stones by using a clinical workstation (Advantage Windows, Version 4.3; GE Healthcare, Waukesha, Wis). Four 5-mm image series were reconstructed (EID, low kilovolt and mixed kilovolt; PCD, bin 1 and T_L) and randomly displayed in the top row of the dual-monitor clinical workstation with 1-mm images randomly distributed across the lower row. Once a stone was seen, a confidence rating for that stone was assigned to each image series as follows: 1, definitely present; 2, probably present; 3, questionable if present; and 4, not seen. Disagreements were recorded and resolved by consensus. Each stone was rated as present if both radiologists gave a detection confidence rating of 1. Confidence ratings of 2–4 were considered as stone not definitively present. Artifacts were recorded.

Stone volume was measured by using in-house software and contrast-to-noise ratio was calculated for all stones rated as definitely present at both EID and PCD CT that were at least 3 mm³ (ie, >10 pixels).

Stone Characterization

Stones were characterized by using the 1-mm low- and high-energy EID images and bin 1 and bin 2 PCD images. An offline workstation (Syngo Multimodality Workplace running Syngo Dual Energy; Siemens Healthcare) segmented and characterized stones as uric acid or non–uric acid. For PCD analysis, a proprietary look-up table was used, which was supplied by the manufacturer. Stones not automatically segmented and characterized were manually measured for greatest linear dimension.

Statistical Analysis

Descriptive statistics of median and range, or mean ± standard deviation for continuous variables, were determined and odds ratios for image thickness preferences were computed. Because we included only participants with stones (20), we calculated the odds of a stone being definitely present with McNemar odds ratio, including 95% confidence intervals as a metric of noninferiority. McNemar odds ratios were compared with those for EID mixed-kilovolt images, with significant differences in stone detection shown by the 95% confidence intervals of the odds ratio not including 1.0. Interrater agreement for stone characterization was performed by using Cohen k coefficient (20). For stone characterization, McNemar odds ratios were calculated for every pair of possible characterizations (eg, uric acid vs non–uric acid or mixed stone). All analyses were performed by using statistical software (SAS version 9.3; SAS, Cary, NC).

Results

Thirty participants (10 women and 20 men; mean age, 61 years; age range, 31–94 years) underwent clinical EID scanning followed by research PCD CT scanning (Table 1). Volumetric CT dose index was similar (EID, 13.9 mGy; PCD, 14.6 mGy; P = .16).

Table 1:

Energy-integrating Detector and Photon-counting Detector CT Settings

Note.—There were 30 participants total, and data in parentheses are the number of participants scanned by using each energy-integrating detector tube potential pair. The mean lateral width of the participants was 36.7 cm ± 5.5. EID = energy-integrating detector, PCD = photon-counting detector.

*Data are mean ± standard deviation.

We found 160 stones that met the criteria for definitely present (confidence rating, 1); all participants had at least one definite stone. A solitary 1.0-mm stone was rated as probably present (confidence rating, 2). No images received confidence ratings of 3 or 4. Most stones (91 of 160; 56.9%) were small, with maximum axial diameter of 3 mm or less (Table 2). Stones were more likely to receive a diagnostic confidence score of 1 (ie, definitely present) for 1-mm images compared with 5-mm images (EID mixed kilovolt odds ratio, 24.4 [95% confidence interval: 3.1, 194.6], P = .003; PCD T_L odds ratio, 48.7 [95% confidence interval: 2.8, 860.5], P = .008; EID low kilovolt odds ratio, 34.3 [95% confidence interval: 4.4, 269.4], P = .001; PCD bin 1 odds ratio, 18.0 [95% confidence interval: 2.2, 146.1], P = .007).

Table 2:

Number of Stones Visually Detected and Characterized According to Confidence of Detection by Using 1-mm Images

Note.—Data are numbers; data in parentheses are percentage and data in brackets are 95% confidence intervals pertaining to the percentages. One-millimeter images were used for all conditions. The odds of detecting a definitely present stone with photon-counting detector low-energy threshold are 1.29 times that when using energy-integrating detector mixed kilovolt images. EID = energy-integrating detector, PCD = photon-counting detector, T_L = low-energy threshold.

*Data in parentheses are 95% confidence interval.

Visual Stone Detection

For stones rated definitely present, PCD and EID yielded similar stone detection by using 1-mm images (PCD, 150 of 160 stones [93.8%]; EID, 148 of 160 stones [92.5%], P = .74; Table 2). By using the 1-mm EID mixed-kilovolt images as a reference for all stones and for small stones (≤3 mm), the McNemar odds ratios for EID low-kilovolt images were 0.70 and 0.78 for all stones and small stones, respectively. The odds of PCD CT T_L helping to detect a definitely present stone were 1.29 times that of EID mixed-kilovolt images for all stones and small stones, with the 95% confidence intervals including 1.0; and for PCD bin 1 images the McNemar odds ratios were 1.25 and 1.67 for all stones and small stones, respectively (Table 2, Fig 2).

Figure 2:

Energy-integrating detector (EID) and photon-counting detector (PCD) scans in a 67-year-old male study participant with known urolithiasis. A small stone (<3 mm; arrow) in the left kidney was detected with questionable confidence (rating of 3) on the, A, EID 1-mm mixed-kilovolt image (enlarged in B). The same stone was visually detected with high confidence (rating of 1) on the, C, PCD low-energy threshold image at the same image thickness (enlarged in D).

Artifacts affected confidence in 1.9% (three of 160) of cases in the PCD bin 1 images and 1.2% (two of 160) of the PCD T_L images, and were related to patient motion.

Stone Characterization

Table 3 shows how the commercial software characterized the stones by using PCD CT and EID CT images. The PCD CT helped to characterize 70.0% (112 of 160) of the stones, including all characterized by the EID system. EID characterized only 54.4% (87 of 160). Of the 25 uncharacterized EID stones that were characterized at PCD CT, about two-thirds were characterized at PCD CT as non–uric acid and one-third were characterized as uric acid (Table 3). Of the stones characterized at EID, 93.1% (81 of 87) were characterized identically at PCD CT. Because of substantial agreement regarding stone type (k = 0.65), stone size (k = 0.67), and the small number of discordant characterizations, we could not reliably estimate differences in stone characterization.

Table 3:

Automatic Renal Stone Characterization at EID and PCD CT

Note.—Data in parentheses are percent. EID = energy-integrating detector, PCD = photon-counting detector, UA = uric acid.

*Stones characterized at EID but not PCD.

^†Stones characterized at PCD but not EID.

^‡Correlation for renal stone characterization between PCD and EID.

PCD CT helped to characterize nearly twice as many small stones as did EID (45 of 91 [49.5%] vs 25 of 91 [27.5%]; P = .002; Table 3; Fig 3).

Figure 3:

Renal stone (arrow) in a 48 year-old female participant in the lower left renal pole detected with confidence of 1 in the EID, A, mixed-kilovolt and, B, low-kilovolt images, and in the, D, PCD T_L and, E, bin 1 images. The stone was not automatically segmented and characterized by using the, C, EID images, whereas the PCD images allowed for automatic segmentation and characterization, which showed, F, a non–uric acid composition.

Of 117 stones rated as definitely present on both CT systems, 83 had volumes of 3 mm³ or greater. For these, the mean contrast-to-noise ratio for the PCD T_L images was similar to that for EID mixed-kilovolt images (17.1 ± 8.5 vs 17.3 ± 10.7, respectively; P = .71), but was significantly higher for PCD bin 1 images compared with EID low-kilovolt images (16.0 ± 8.2 vs 14.1 ± 8.6, respectively; P < .001).

Discussion

Our results demonstrated similar reader confidence in renal stone detection by using a whole-body PCD CT system compared with an EID dual-source dual-energy CT system. PCD CT permitted automatic segmentation and characterization of a greater number of small stones. These findings show the potential for PCD CT to overcome the current limitation of commercial dual-energy CT in automatic characterization of small stones. Despite large confidence intervals, the McNemar odds ratios confirmed that both the T_L PCD images, created by using all photons between 25 and 140 keV, and the bin 1 PCD images, created by using lower energy photons between 25 and 75 keV, are not inferior to the corresponding EID data in detecting renal stones.

CT number ratio thresholds to differentiate uric acid from non–uric acid stones depend on several factors, including tube voltage difference, use of a tin filter, patient size, and image noise (4,21). An ex vivo characterization of PCD CT performance to differentiate stone composition reported that PCD CT outperformed second-generation dual-source dual-energy CT for larger patients (18). Because the same tube potential and energy bins are used for all patient sizes with the multienergy PCD CT system, patient size does not affect characterization performance. The in vivo results in this work are consistent with the ex vivo results.

Previous literature (22,23) has shown that PCD CT images show lower noise if the reconstruction kernel is matched. We found that stone contrast-to-noise ratio was similar for EID mixed kilovolt and PCD T_L images, with improved contrast-to-noise ratio for PCD bin 1 compared with EID low-kilovolt images.

The PCD system used in our study is a research system with the same x-ray tube, antiscatter grid spacing, focal spot size, magnification, and other components as the EID system used in this work so that the effect of the PCD itself on stone detection and characterization could be studied. Of note, because of lack of automatic exposure control on the PCD system, the tube current is not elevated for lateral projections as it is on the EID CT system so that matching volumetric CT dose index actually penalizes the PCD system. Our study did not take this into account. Investigators have previously demonstrated no significant difference in image quality when comparing the macro mode of the PCD system without automatic exposure control (used herein) to a conventional CT system with automatic exposure control by using a similar volumetric CT dose index–matching technique (14).

Our study had several limitations. The subject sample size was limited; however, there were a large number of total stones. Stone presence and mineral type were determined on the basis of the 1-mm CT data without external validation. We treated stones within participants as independent, but there may be correlations between stones within the same participant that might affect their detectability or characterization. Because we included participants who underwent dual-energy CT stone characterization, all patients had stones, precluding traditional noninferiority comparisons (20). We consequently used McNemar odds ratios and 95% confidence intervals to compare the performance of the PCD system to the EID system.

In conclusion, the evaluated PCD CT system has the potential to improve the automatic characterization of small stones (≤3 mm). This diagnostic task might be especially important for patients with chronic nephrolithiasis, in whom small stones in the calyces may be the source of emerging stones. Detection of these nascent stones could enable clinical intervention, such as a change of diet or addition of a medication, to prevent further stone growth.

Summary

The evaluated photon-counting–detector CT system has the potential to improve the automatic characterization of small stones (≤3 mm).

Implication for Patient Care

• ■ Photon-counting CT may help to provide more reliable detection and automatic characterization of renal stones than conventional CT.