Comparing the clinical utility of single-shot, readout-segmented and zoomit echo-planar imaging in diffusion-weighted imaging of the kidney at 3 T

We compared the clinical utility of single-shot echo-planar imaging (SS-EPI) using different breathing schemes, readout-segmented EPI and zoomit EPI in the repeatability of apparent diffusion coefficient (ADC) measurements, cortico-medullary contrast to noise ratio (c-mCNR) and image quality. In this institutional review board-approved prospective study, some common clinically applicable diffusion-weighted imaging (b = 50, 400, 800 s/mm2) of kidney on 3.0 T MRI were performed on 22 volunteers using SS-EPI with breath-hold diffusion-weighted imaging (BH-DWI), free-breathing (FB-DWI), navigator-triggered (NT-DWI) and respiratory-triggered (RT-DWI), readout-segmented DWI (RS-DWI), and Zoomit DWI (Z-DWI). ADC and c-mCNR were measured in 12 anatomic locations (the upper, middle, and lower pole of the renal cortex and medulla), and image quality was assessed on these DWI sequences. A DWI with the optimal clinical utility was decided by systematically assessing the ADC repeatability, c-mCNR and image quality among the DWIs. For ADC measurements, Z-DWI had an excellent intra-observer agreement (intra-class correlation coefficients (ICCs): 0.876–0.944) and good inter-observer agreement (inter-class ICCs: 0.798–0.856) in six DWI sequences. Z-DWI had the highest ADC repeatability in most of the 12 anatomic locations of the kidneys (mean ADC absolute difference: 0.070–0.111 × 10−3 mm2/s, limit of agreement: 0.031–0.056 × 10−3 mm2/s). In all DWIs, Z-DWI yielded a slightly higher c-mCNR than other DWIs in most representative locations (P > 0.05), which was significantly higher than BH-DWI and FB-DWI in the middle pole of both kidneys and the upper pole of the left kidney (P < 0.05). In addition, Z-DWI yielded image quality that was similar to RT-DWI and NT-DWI (P > 0.05) and superior to BH-DWI, FB-DWI and RS-DWI (P < 0.05). Our results suggest that Z-DWI provides the highest ADC reproducibility, better c-mCNR and good image quality on 3.0 T MRI, making it the recommended sequence for clinical DWI of the kidney.


Measurement of cortico-medullary contrast to noise ratio(c-mCNR).
For the measurement of c-mCNR, a good agreement between reader 1 and reader 2 was found in the upper pole (RK: r = 0.779; LK: r = 0.891), middle pole (RK: r = 0.775; LK: r = 0.818) and lower pole (RK: r = 0.72; LK: r = 0.854) with Z-DWI (Tables 3 and 4). Furthermore, the Z-DWI has a slightly higher c-mCNR than other DWIs in most representative Table 1. The ADC measurement in six DWI techniques and their Intra-and Interobserver agreement between them. ADC are given in *10 −3 mm 2 /s. Mean ADC values measured in different anatomical regions. Data in parentheses are 95% confidence intervals. P values were gained by using the paired t test to compare differences for reader1 between the first and second ADC measurement, and to compare for the first ADC measurement between reader1 and reader2. P < 0.05 were considered significant difference (*). LK left kidney, RK right kidney, CV coefficient of variation, BH-DWI breath-hold DWI, FB-DWI free-breathing DWI, NT-DWI navigator-triggered DWI, RS-DWI readout-segmented DWI, RT-DWI respiratory-triggered DWI and Z-DWI Zoomit DWI.  www.nature.com/scientificreports/ Image quality analysis. The two readers had an excellent agreement in evaluating the five aspects (K1-K5) of image quality (Kappa value 0.945-0.989). Z-DWI had a high score in terms of image blurring (5 points), severity of artifacts (4 points), sharpness of boundaries (5 points), clarity of the renal cortex and medulla (5 points), and overall image quality (5 points), which was similar with the image quality of RT-DWI and NT-DWI (P > 0.05). However, Z-DWI had a better image quality than BH-DWI in K4 (ADC map) (P < 0.05), FB-DWI in K2 (all P < 0.05), K4 and K5 (ADC map) (all P < 0.05), and RS-DWI in all image quality aspects except for K2 and K4 (ADC map) (all P < 0.05) (

Discussion
Currently, BH-DWI, FB-DWI, NT-DWI, RS-DWI, RT-DWI and Z-DWI have widely been used for the diagnosis of renal diseases and evaluate renal function 3,6,8,12,13 . For these DWIs, the reliability of ADC value and good image quality are vital in detecting renal disease and assessing renal function accurately. To our knowledge, this is the first MRI study to compare these DWIs systematically, by evaluating the intra-and inter-observer agreement in ADC measurements, reproducibility of ADC values and image quality to establish the most reliable clinically applicable renal DWI sequence. ADC values derived from coronal renal DWI exhibited moderate-to-good agreement to axial DWI 11 . In our study, coronal renal DWI was performed because it can provide full coverage of the kidney shape. The mean ADC value in the renal cortex falls between 1.429 and 2.082 × 10 −3 mm 2 /s and 1.211-1.749 × 10 −3 mm 2 /s in the medulla in 12 representative sections (the upper, middle, and lower pole of both kidneys), which were near the lower limit of the values reported in the literature ((1.78 ± 0.11) × 10 −3 mm 2 /s for the renal cortex and (1.48 ± 0.13) × 10 −3 mm 2 /s for renal medulla in healthy volunteers 21 . Furthermore, our results showed that the 95% CI of ADC measurements in the cortex was higher than that in the medulla using the six DWI sequences, consistent with Sulkowska et al. 22 . Previous ADC values obtained with NT-DWI 10 and RS-DWI 23 were similar to our findings but were slightly higher with BH-DWI 10 and Z-DWI 13 than that in our results. In our results, Z-DWI yielded lower ADC values in the cortex and the medulla than the other five DWIs, which is consistent with Cai et al. findings that showed that the mean tumor ADC values of rFOV-DWI were significantly lower than those of fFOV-DWI (1.237 ± 0.228 × 10 −3 mm 2 /s vs 1.683 ± 0.322 × 10 −3 mm 2 /s, P < 0.001) in patients with gastric   Table 4. Cortico-medullary contrast to noise ratio (c-mCNR) of left kidney. P < 0.05 were considered significant difference (*). P values were presented only when they were less than 0.05. c-mCNR (Corticomedullary contrast to noise ratio) was measured in different anatomical regions with b = 800 s/mm 2   www.nature.com/scientificreports/ cancer 24,25 . The possible reason was that DWI with reduced FOV produce images with sharper margins and anatomic structural visualization 26 , which is helpful in drawing ROI in renal the cortex and medulla, yielding a stable and low ADC value. This suggests that a lower ADC value should be used in clinical work when using Z-DWI. In addition, Z-DWI has the best intra-observer agreement (intra-class ICCs: 0.906-0.944) and inter-observer agreement (inter-class ICCs: 0.798-0.856) among the six sequences, indicating that Z-DWI is sufficiently reliable and repeatable when assessing ADC measurements. The possible reason for this result is that a reduced ("zoomed") FOV in the phase-encoding direction decreases the influence of gastrointestinal peristalsis and respiratory motion artifacts on kidney images. In addition, RT-DWI and NT-DWI can reduce the influence of motion artifacts by respiratory-and navigator-triggered techniques. However, it is at the cost of rather long and uncertain scan times (more than 120 s in both sequences), which can markedly increase patients' discomfort and sensitivity to motion 27 . Consequently, the intra-and inter-observer agreements with RT-DWI and NT-DWI were lower than with Z-DWI. Previous studies have shown that Z-DWI has obvious advantages in cervical cancer 28 , thyroid micronodules 29 , cervical spinal cord 30 , etc. It enables clearer identification of lesions and reduction of image artifacts. Our research has further verified its value in kidney applications. Moreover, we found that the CV was less than 3% in all measurements, suggesting that the ADC measurements were reliable and consistent in all DWIs.
Our results indicate that Z-DWI has the best ADC repeatability because it yielded the least mean absolute differences of ADCs and LOAs in all the anatomical sections. This finding may be related to the "zoomed" technique in the direction of phase-encoding, which, when combined with dynamic, spatially selective RF pulses,  Kappa  value  R3  R4  R3  R4  R3  R4  R3  R4  R3  R4  R3  R4 Overall Pairwise    The image quality of Z-DWI was significantly different from BH-DWI, FB-DWI and RS-DWI in the three representative section (n-2 slice, n slice and n + 2 slice) in clarity of the renal cortex and medulla (ADC map) (all P < 0.05). Z-DWI was slightly superior to RT-DWI and NT-DWI in the three representative section (n-2 slice, n slice and n + 2 slice) in sharpness of boundaries (ADC map); however, the difference in image quality between the three was also not significant (all P > 0.05). www.nature.com/scientificreports/ further improved image quality in renal imaging considerably more than other DWIs 15,31 . According to previous studies of abdominal organs, different breathing schemes will affect the absolute ADC value 32 . The study of Yıldırım İO et al. 33 found that compared with conventional DWI sequences, Z-DWI may be more effective in the diagnosis and monitoring of treatment and postoperative responses in patients with varicocele. Therefore, the good repeatability of Z-DWI helps us to evaluate the ADC value of renal disease quantitatively. Our study verifies that Z-DWI has the best consistency and reproducibility, which is of great significance to the future clinical applications of renal DWI sequences. Our results also showed that all the LOAs were around 20-30% of the mean ADC values. This is in line with previous studies that recommended at least a 30% change in ADC values when evaluating a lesion's response to treatment with the same DWI technique 6,23 .
The DWIs with a long scan time (like Z-DWI, RS-DWI, RT-DWI and NT-DWI) can reduce the artifacts in DWI protocols, but this in turn can markedly increase patient's discomfort and decrease image quality. In our study, Z-DWI yielded a high score in terms of imaging blurring, sharpness of boundaries, clarity of the renal cortex and medulla, and overall image quality, which has the similar image quality to RT-DWI and NT-DWI (P > 0.05) and superior to RS-DWI (P < 0.05). The possible reason is that the "zoomed" technique in the direction of phase-encoding, combined with dynamic, spatially selective RF pulses reduced susceptibility artifacts markedly and gained considerable image quality improvements in renal imaging 15,31 . Although RS-DWI can reduce T2 blurring and susceptibility effects 12 , its long acquisition time (226-379 s in our study) makes it prone to motion artifacts, reducing the image quality, especially for the mobile kidney. Furthermore, Z-DWI had a higher score than BH-DWI in clarity of the renal cortex and medulla (ADC map, P < 0.05) and RS-DWI in clarity of the renal cortex and medulla (all P < 0.05). This indicates that imaging with Z-DWI provides a clearer margin between the renal cortex and medulla and helps to locate the orientation of renal lesions and precisely measure ADC value in the cortex and medulla. In addition, Z-DWI was better than FB-DWI and RS-DWI in severity of artifacts (P < 0.05), which is similar to a previous study where FB-DWI and RS-DWI had more artifacts compared to Z-DWI 13,36 .
This study also has some limitations. First, the volunteers included in this study are all young with better breathing coordination, which is somewhat different from the clinical situation of patients with kidney disease. Secondly, this study was performed in normal kidneys, without any lesions, to ensure the same condition of the kidney to avoid the bias of ADC measurements due to inhomogeneity that lesions might cause. Finally, in order to evaluate the image quality of different kidney regions, this study uses a coronal scan, which increases the impact of respiratory motion artifacts on the image.
In summary, Z-DWI had an excellent intra-observer agreement and good inter-observer agreement among the six sequences. Furthermore, Z-DWI had the highest ADC repeatability and c-mCNR in most of the 12 locations of the kidneys observed. In addition, Z-DWI had a similar image quality with RT-DWI and NT-DWI and better image quality than BH-DWI, FB-DWI and RS-DWI (P < 0.05). Therefore, Z-DWI is the optimal renal DWI sequence that can be used as a reliable quantitative parameter and therapeutic biomarker for patients with renal disease and evaluation of renal function. Thus, it is recommended as the DWI sequence for clinical examination of the kidney due to its good image quality and reliable diagnostic confidence.

Materials and methods
Ethics statement and participants' enrollment. This prospective study was approved by the research ethics committee of our institution (Xiangya Hospital, Central South University, China). The authors confirm all data has informed written informed consent obtained from each participant. All methods were performed in accordance with the relevant guidelines and regulations and strictly abide by the Declaration of Helsinki. 22 healthy young volunteers with similar age (juniors in a medical college, mean: 21 years, range: 20-22 years) were enrolled (12 males, 10 females).
In our study, the inclusion criteria included: (a) no history of albuminuria, hematuria and weight loss; (b) no history of any kidney surgery; (c) ability of the subject to hold his or her breath for up to 20 seconds. The exclusion criteria included: (a) contraindications to MR imaging; (b) history of any kidney disease and surgery.
MR imaging protocol. Magnetic resonance examinations were performed on a 3.0 T system (MAG-NETOM Prisma, Siemens Healthcare, Erlangen, Germany) with an 18-channel anterior surface body coil combined with 12 elements of a 32-channel spine coil. Each subject was scanned twice in the DWI series. The DWI series included end-expiratory breath-hold DWI (BH-DWI) (one breath-hold), free-breathing DWI (FB-DWI), navigator-triggered DWI (NT-DWI), readout-segmented DWI (RS-DWI)(with respiratory-triggering), respiratory-triggered DWI (RT-DWI) and Zoomit DWI (Z-DWI) (with respiratory-triggering). Three b values of 50, 400, and 800 s/mm 2 were sampled in three orthogonal diffusion directions (three-scan trace) for all DWIs. A 5 min rest was allowed between two identical sessions. The scan parameters were kept as close as possible, and the detailed parameters of all sequences are summarized in Table 6. The imaging parameters of the two scans were consistent. Each participant had 12 scans (six scans using the 6 techniques in each session). The fat suppression was achieved with spectral adiabatic inversion recovery in all DWI sequences, and the acceleration factors were 2 in all sequences. A k-space-based parallel imaging technique was used. The scan time was recorded. ADC value measurement and repeatability evaluation. 12 ROIs were drawn on the b = 50 s/mm 2 images, including the upper, middle and lower poles of cortex and medulla on both kidneys. ROIs 20-24mm 2 in size 37,38 were positioned the on the b = 50 s/mm 2 image (Fig. 5A,B), and then copied to the ADC map for ADC measurements (Fig. 5C) and b = 800 s/mm 2 images for c-mCNR measurements (Fig. 5D). Then, ROIs were drawn in the second scan and in the repeated series of the other five sequences in a similar manner. The ADC measurements were repeated one week after the first measurement to avoid recall bias. The second radiologist repeated the same measurement. The ADC value were gained using the following formula by the log-linear fitting algorithm with three different b factors (b = 50, 400, 800 s/mm 2 ): I1, I2, and I3 are the measured diffusion-weighted images in three orthogonal gradient directions, and D1, D2 and D3 are the corresponding diffusion coefficients.
In addition, CV of ADC value was used to assess the relative degree of dispersion between ADC value measurements, which was calculated as the following formula: Here, SD was the standard deviation of ADC value and ADC Mean was the mean value of ADC value in various representive point 39 . It indicated that the ADC measurement was reliable when CV less than 0.15.
Cortico-medullary contrast to noise ratio (c-mCNR). The signal intensity (SI) was measured in different anatomical regions with b = 800 s/mm 2 images, including the upper, middle and lower poles of cortex and medulla (1) www.nature.com/scientificreports/ on both kidneys. Moreover, the background signal standard deviation (SD) measured using an equally sized ROI placed at a nearby background (air) in the corresponding section, close to the site of the kidney ROI, and avoiding any prominent artifacts (Fig. 5D). The following formula was used to calculate the corresponding c-mCNR of different DWI sequences: where SI cortex and SI medulla were the signal intensity of the specific position ROI (for instance, the upper pole of right kidney). SD background was the standard deviation of the chosen artifact-free ROI positioned on the background (air) of the corresponding slice. In all volunteers, CNR were measured once in 1 week by reader 1 and reader 2. Mean values of c-mCNR with the standard deviation and 95% confidence interval were recorded.
The evaluation of image quality. The image quality of the six DWIs on the ADC map and DWI images at b = 50, 400 and 800 s/mm 2 were evaluated by two radiologists (reader 3 and reader 4), respectively. The score criteria of image quality for each DWI are shown in Table 7.
Statistical analysis. The mean value and standard deviation (SD) of ADC values of 12 ROIs in cortex and medulla on both kidneys were used to estimate the consistency of ADC measurement. The t-test was used to compare the difference between the first and second readers' measurements (inter-observer agreement) and the difference between repeated measurements (intra-observer agreement). The intra-and inter-class correlation coefficients (ICCs) (and 95% confidence intervals) were used to evaluate the intra-and inter-observer agreement, respectively. An ICC value greater than 0.70 indicates good consistency. In order to evaluate the repeatability of ADC, we used the Bland-Altman method, which compares the 95% confidence interval (limit of agreement [LOAs]) between the first and second sets of DWI sequences and the mean absolute difference of ADC values. The median of image quality evaluation was obtained from the 3 b-value DWI images and ADC map. The inter-observer agreement for image quality was analyzed by calculating weighted kappa coefficients (quadratic weighting), with kappa values of 0.01-0.25 representing slight agreement, 0.25-0.45 fair, 0.45-0.65 moderate, 0.65-0.85 substantial, and 0.85-1.00 almost perfect agreement. The Friedman test was (3) c − mCNR = |SI cortex − SI medulla | SD background Figure 5. Schematic diagram of typical ROI placement in the renal cortex and medulla. Raw diffusionweighted imaging (DWI) at b = 50 s/mm 2 (A), six representative ROIs (3 ROIs each for the renal cortex and medulla in superior, middle and inferior zones, respectively) on DWI at b = 50 s/mm 2 (B), corresponding ADC map (C) and b = 800 s/mm 2 image (D) for c-mCNR measurements. First, the DWI slice (using b = 50 s/ mm 2 ) with the largest renal section was chosen and a straight line was drawn along the upper and lower poles of the kidney (white dashed line). Then, a perpendicular bisector was drawn (white dashed line). Second, the medullary zone adjacent to the white dashed, which has a clear lower signal intensity was identified and the ROIs representing the superior, middle and inferior zones were drawn manually. Subsequently, three similar ROIs were drawn in renal cortex based on the representative medulla positions. These ROIs were copied to the corresponding ADC map for ADC measurement. Moreover, the background signal standard deviation (SD) for c-mCNR measured using an equally sized ROI placed at a nearby background (air) in the corresponding section, close to the site of the kidney ROI, and avoiding any prominent artifacts. www.nature.com/scientificreports/ used to compare the differences between the six methods, and the Dunn-Bonferroni post-hoc test adjusted for all significant pairwise comparisons. Statistical analysis was performed using SPSS (version 19.0, Chicago, IL) software. When the P-value was less than 0.05, the difference is considered significant.

Data availability
All data generated or analysed during this study are included in this published article and its supplementary information files.