Single-Breath-Hold MRI-SPACE Cholangiopancreatography with Compressed Sensing versus Conventional Respiratory-Triggered MRI-SPACE Cholangiopancreatography at 3Tesla: Comparison of Image Quality and Diagnostic Confidence

To compare two magnetic resonance cholangiopancreatography (MRCP) sequences at 3 Tesla (3T): the conventional 3D Respiratory-Triggered SPACE sequence (RT-MRCP) and a prototype 3D Compressed-Sensing Breath-Hold SPACE sequence (CS-BH-MRCP), in terms of qualitative and quantitative image quality and radiologist’s diagnostic confidence for detecting common bile duct (CBD) lithiasis, biliary anastomosis stenosis in liver-transplant recipients, and communication of pancreatic cyst with the main pancreatic duct (MPD). Sixty-eight patients with suspicion of choledocholithiasis or biliary anastomosis stenosis after liver transplant, or branch-duct intraductal papillary mucinous neoplasm of the pancreas (BD-IPMN), were included. The relative CBD to peri-biliary tissues (PBT) contrast ratio (CR) was assessed. Overall image quality, presence of artefacts, background noise suppression and the visualization of 12 separated segments of the pancreatic and bile ducts were evaluated by two observers working independently on a five-point scale. Diagnostic confidence was scored on a 1–3 scale. The CS-BH-MRCP presented significantly better CRs (p < 0.0001), image quality (p = 0.004), background noise suppression (p = 0.011), fewer artefacts (p = 0.004) and better visualization of pancreatic and bile ducts segments with the exception of the proximal CBD (p = 0.054), cystic duct confluence (p = 0.459), the four secondary intrahepatic bile ducts, and central part of the MPD (p = 0.885) for which no significant differences were found. Overall, diagnostic confidence was significantly better with the CS-BH-MRCP sequence for both readers (p = 0.038 and p = 0.038, respectively). This study shows that the CS-BH-MRCP sequence presents overall better image quality and bile and pancreatic ducts visualization compared to the conventional RT-MRCP sequence at 3T.


Introduction
Magnetic resonance cholangiopancreatography (MRCP) is a non-invasive imaging method used in everyday clinical practice to assess anatomical features and abnormalities of the intrahepatic and extrahepatic bile ducts and of the pancreatic ducts [1][2][3][4][5]. MRCP does not involve the risks associated with endoscopic retrograde cholangiopancreatography (ERCP) such as acute pancreatitis, bowel perforation, infections, and bleeding [6]. Recent years have witnessed the development of three-dimensional (3D) imaging, which provides better image quality and greater diagnostic confidence compared to two-dimensional (2D) imaging [7].
Two 3D MRCP breathing management methods are currently available. The freebreathing (FB) method relies on respiratory gating has well-established diagnostic performance and spatial resolution characteristics when used to evaluate diseases of the bile ducts and pancreas [2]. However, this method requires quite long acquisition time. The other method, in which the images are acquired during a single breath-hold (BH), is being actively developed.
Research into means of improving MRCP sequences has several goals, of which the most important is reduction in the acquisition time in order to limit motion artifacts, notably those due to breathing, which are particularly challenging when imaging the abdomen. A shorter acquisition time also improves the comfort of the patient, who must remain immobile throughout the acquisition. Another advantage is the imaging of a greater number of patients during a given time, which decreases costs and wait-list times.
In practice, the acquisition time and diagnostic quality of FB MRCP with respiratory triggering (RT) are variable and difficult to predict. Only part of the k-space is acquired during each breathing cycle. Diagnostic performance may be adversely affected by irregular breathing due, for instance, to abdominal pain or failure of the machine to detect breaths. However, irregular breathing can also occur in patients with a World Health Organization (WHO) performance status of 0 for unknown reasons [8]. FB with RT acquisition times of 7 min [9], and 6 min [10], have been reported. Longer acquisition times are often associated with poorer image quality [9,11].
3D MRCP is obtained using a T2-weighted fast spin-echo sequence with the variableflip-angle technique, such as SPACE (Sampling Perfection with Application-optimized Contrast using different flip-angle Evolutions), CUBE, and VISTA (Volume Isotropic Turbo spin echo Acquisition). The high flip angles of radiofrequency pulses for the lines near the center of the Fourier space ensure a good contrast-to-noise ratio (CNR), while the lower angles for the lines at the periphery of the Fourier space provide good spatial resolution. The clinical relevance of the SPACE technique for MRCP has been demonstrated [12].
Methods that have been evaluated to decrease the acquisition time include parallel imaging (PI) and compressed sensing (CS), which combines data undersampling with iterative reconstruction. PI has entered the mainstream of clinical practice. The decrease in acquisition time is achieved by undersampling the k-space; for instance, by skipping every other line. The main drawbacks are the moderate acceleration factors in the 2-4 range and a decrease in the CNR [13,14]. CS has undergone considerable development in recent years as a method for imaging moving targets, such as the heart and abdominal organs [15][16][17][18][19]. CS relies on the sparsity of redundant data within standard magnetic resonance images. Random undersampling of the compressible representation of an image is performed, and the image is then restored by iterative reconstruction [20].
Several studies have evaluated the clinical feasibility of CS with FB or BH acquisition, with interesting results that favored BH sequences [9,[21][22][23]. A prototype CS-BH-SPACE MRCP sequence produced by Siemens Healthcare (Erlangen, Germany) seeks to eliminate motion artifacts during a BH.
The objective of this study was to prospectively compare two MRCP sequences at 3T, namely, the conventional 3D RT-SPACE sequence (designated RT-MRCP hereafter) and the prototype 3D CS-BH-SPACE sequence (designated CS-BH-MRCP hereafter), in terms of qualitative and quantitative image quality and radiologist's diagnostic confidence for detecting common bile duct (CBD) lithiasis, biliary anastomosis stenosis in liver-transplant recipients, and communication of pancreatic cyst with the main pancreatic duct (MPD), the latter allowing the non-invasive diagnosis of branch-duct intraductal papillary mucinous neoplasm of the pancreas (BD-IPMN).

Study Population
This prospective single-center study included patients who underwent 3T MRCP between September 2018 and August 2020 in our institution. Consecutive patients older than 18 years of age who had clinical and laboratory-test abnormalities suggestive of choledocholithiasis or biliary anastomosis stenosis after liver transplant, or required MR imaging for cystic pancreatic lesion characterization, were included. Exclusion criteria were contraindications to MRI, moderate-to-abundant ascites, an inability to maintain a breath-hold. Informed consent was obtained from every subject.

MRI Protocols
A 3T machine (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany) was used to acquire all MRCPs. An 18-channel body and 32-channel spine matrix coils were used for signal reception. The main advantage of 3T machines is a better CNR, which decreases the acquisition time and improves spatial resolution [22]. A recent study comparing 1.5T and 3T machines showed that image quality was better with CS than with the conventional sequence using the 3T machine, but that image quality was better with the conventional sequence using the 1.5T machine [24]. The patients started fasting 4 h before the scan; they were positioned supine and entered into the tunnel feet first. Before entering the tunnel, patients without diabetes drank a glass of pineapple juice, whose high manganese content decreases the T2 relaxation time of fluids, thereby suppressing the signal from the proximal digestive structures and limiting background noise [25]. SPACE was used to acquire 3D images. Conventional free-breathing with a navigator triggering sequence (RT-MRCP), that is the standard 3D-MRCP sequence used in our department, and the prototype BH sequence with compressed sensing (CS-BH-MRCP), were acquired according to the acquisition parameters shown in Table 1. For each MRCP sequence, maximum-intensity projection (MIP) images were generated automatically and used in addition of the native images for image analysis. The standard MRCP protocol also included a coronal and axial T2 Half-Fourier Acquisition Single-Shot Turbo Spin Echo (HASTE) sequence, as well as an axial T1 Volumetric Interpolated BH Examination (VIBE) with fat-saturation sequence. RT-MRCP acquisition times were recorded. The time required for CS-BH-MRCP sequence image reconstruction on the MRI console was about 10 min. No other sequence could be acquired during reconstruction, and this time was therefore used to remove the patient from the machine and to prepare the next patient. Consequently, image quality could not be evaluated before the patient had left the MRI machine.

Image Quality Analysis
All quantitative and qualitative image analyses were performed on an image processing console with picture archiving and communication system (PACS; Centricity, GE Healthcare, Chicago, IL, USA).

Quantitative Image Evaluation
Conventional methods based on the background region-of-interest (ROI), that are commonly used for signal-to-noise ratio and CNR assessment [12], might be unreliable with sequences using undersampling methods, such as PI and CS, due to the heterogeneous signal intensity of the background [26][27][28]. As it has been previously reported, the relative CBD to peri-biliary tissues (PBT) contrast ratio (CR) was assessed instead of the conventional SNR and CNR [23].
For each patient and each sequence, a representative section of the CBD was selected. Then, circular regions of interest (ROIs) were traced on the CBD and PBT and the mean signal intensities were recorded. The CBD ROI was at least 5 mm 2 and was placed in a uniform artifact-free region in the middle of the duct. A similar ROI was positioned on the peri-CBD tissues, avoiding artifacts and fluid-containing structures. Note that the ROIs were drawn on the native images. Figure 1 shows the ROIs placement. The CR was estimated using the following formula [23,29,30]: where CR is the estimated contrast ratio, SI CBD and SI PBT are the mean signal intensities of the CBD and the PBT, respectively.

Qualitative Image Evaluation and Diagnostic Confidence
The images of the two MRCP sequences were read by two observers working independently of each other, a senior radiologist and a junior radiologist with 9 and 4 years of experience reading MRCPs, respectively. The readers were aware of the reason for per-forming MRCP but not of the MRCP protocol used.
Both the native images and the MIP images were used for the analysis. The imageprocessing console enabled multiplanar reconstructions. The other sequences from the standard protocol previously cited were also systematically reviewed.
Overall image quality, presence of artefacts, and background noise suppression were evaluated on a five-point Likert scale (Table 2). When artefacts were seen, their type was recorded as blurring artefacts, metal-induced artefacts, or wrap-around artefacts. The readers also assessed the visualization of 12 segments of the pancreatic and bile ducts: the proximal and distal parts of the CBD; the confluence of the cystic duct and CBD; the proximal, central, and distal parts of the MPD; the intrahepatic bile ducts (IHBDs) to their primary branches (right and left IHBDs) and secondary branches (right anterior and posterior sectoral ducts and left medial and lateral branches). Each of these 12 segments were evaluated using a five-point Likert scale, with higher scores indicating better duct visualization ( Table 2). Figure 2 shows these segments on an MIP reconstruction of a CS-BH-MRCP image.
In addition, adequate visualization of the entire biliary system as scores of 3 or more for the CBD, cystic duct confluence, and intrahepatic bile ducts to their primary and secondary branches, was defined. Similarly, adequate entire MPD visualization was de-fined as a score of 3 or more for all three segments of the MPD.
Degree of diagnostic confidence was scored on a 1-3 scale, with higher scores indicating greater confidence for the presence or absence of: CBD lithiasis, cyst communication with the MPD, biliary anastomosis stenosis (liver-transplant recipients) ( Table 3). When diagnostic confidence was moderate to definitive (score of 2 or 3), the readers indicated whether CBD lithiasis, cyst communication with the MPD or biliary anastomosis stenosis (liver-transplant recipients) was seen (yes/no). In patients with more than one pancreatic cyst, communication with the MPD was evaluated only for the largest cyst.  CBD lithiasis, pancreatic cyst communication with the MDP and biliary anastomosis stenosis detection, based on the consensus between the two readers (confidence diagnostic of 2 or 3) with both sequences was reported. When available, the findings from ERCP for cholelithiasis were also reported; however, they did not serve as the reference standard, as lithiasis migration might have occurred during the interval between MRCP and ERCP.

Statistical Analysis
Variable values (CR) measured on the RT-MRCP sequence and the CS-BH-MRCT sequence were compared using Wilcoxon's test. Overall image quality, presence of artefacts, background noise suppression, and duct visualization were described as mean ± standard deviation (SD) and range (minimum-maximum). Overall image quality and presence of artefacts for different RT-MRCP acquisition times were compared by applying the Kruskal-Wallis test. Inter-observer agreement was assessed by computing Cohen's κ, which can vary from 0 to 1 (Appendix A, Table A1). The chi-square test was applied to compare visualization of the overall bile duct system and entire MPD. Bowker's test was chosen to compare diagnostic confidence between the two sequences. MacNemar's test was chosen to compare CBD lithiasis detection between the two sequences. Values of p smaller than 0.05 were taken to indicate significant between-group differences. Statistical analyses were performed using Stata 14.0 software (StataCorp, College Station, TX, USA).

Study Population
Of 70 patients who met our inclusion criteria, two were unable to maintain a breathhold, leaving 68 patients for the analysis. Table 4 lists the main patient features. Among them, 12 had history of cholecystectomy. Among the 54 patients referred for suspected choledocholithiasis, all presented liver function test abnormalities, 11 suffered from acute pancreatitis, and 10 suffered from acute cholecystitis. Teen patients were referred for pancreatic cyst characterization with suspected BD-IPMN and four liver-transplant patients were scanned for suspected stenosis of the biliary anastomosis.

Quantitative Image Evaluation
Concerning CBD and PBT CRs, a significant difference was found between the two sequences (p < 0.0001), that favored the CS-BH-MRCP sequence (Table 5). Table 5. Quantitative evaluation: comparison of the common bile duct (CBD) to peri-biliary tissues (PBT) contrast ratios (CR) between the two sequences RT-MRCP and CS-BH-MRCP. RT-MRCP, respiratory-triggered magnetic resonance cholangiopancreatography; CS-BH-MRCP, compressed sensing breath-hold magnetic resonance cholangiopancreatography; CR, contrast ratio; CR = (SI CBD − SI PBT )/(SI CBD + SI PBT ); CBD, common bile duct; PBT, peri-biliary tissues; * range: minimum-maximum values. Table 6 compares the mean values of each variable regarding qualitative evaluation obtained by the two readers for each of the two sequences. Appendix A, Table A2 reports the results of the qualitative evaluation of the two MRCP sequences by each of the two readers. Interobserver agreement for the assessment of image quality qualitative eval-uation variables ranged from moderate (κ = 0.41-0.60) to nearly perfect (κ = 0.81-1.00) (Appendix A, Table A3). Table 6. Qualitative evaluation: Comparison of the two sequences RT-MRCP and CS-BH-MRCP: mean value [mean ± SD, (range)] of each qualitative variable based a 5-point Likert scale, obtained by the two readers.

Overall Image Quality, Artefacts and Background Noise Suppression
The CS-BH-MRCP sequence was associated with significantly better image quality (p = 0.004), significantly fewer artefacts (p = 0.004), and significantly better back-ground noise suppression (p = 0.011) overall. Blurring artefacts occurred with both sequences but were more numerous with the free-breathing sequence. Metal-induced artefacts were seen in two patients: For the first patient, the artefact was due to spinal internal fixation material and occurred on the free-breathing sequence and, for the second, the artefact was due to a weighted nasogastric feeding tube inserted as part of the management of acute pancreatitis and occurred on the BH sequence.

Bile Ducts Visualization
Whereas visualization of the distal CBD was significantly better with the CS-BH-MRCP sequence (p = 0.015), no significant difference was found for the proximal CBD or cystic duct confluence (p = 0.054 and 0.459, respectively). Visualization of the right and left primary IHBD was significantly better with the CS-BH-MRCP sequence (p = 0.022 and 0.018, respectively), whereas no significant difference was found for the secondary IHBDs (Table 6).

Main Pancreatic Duct Visualization
For the distal and proximal parts of the MPD, visualization was significantly better with the CS-BH-MRCP sequence (p = 0.001 and 0.032, respectively), whereas no difference was found for the central part (p = 0.885). The central part was often not seen on the CS-BH-MRCP sequence, as it was outside the acquisition volume.

Entire Biliary System and Pancreatic Duct Visualization
Visualization of the entire biliary system was obtained in 48.5% (33/68) of patients with the RT-MRCP sequence and 35.3% (24/68) of patients with the CS-BH-MRCP se-quence; the difference was not significant (p = 0.118). Visualization of the entire MPD was obtained in 35.3% (24/68) of patients with the RT-MRCP sequence and in 51.5% (35/68) of patients with the CS-BH-MRCP sequence, with no significant difference (p = 0.057). Figure 3 shows example MRCP images obtained with RT-MRCP and CS-BH-MRCP sequences with their qualitative image-quality criteria scorings.  Table A4).

Diagnostic Confidence
Overall, diagnostic confidence was significantly better with the CS-BH-MRCP sequence for both readers (p = 0.038 for senior and p = 0.038 for junior, Appendix A, Table A5). Cross-tabulation analyses in Table 7 shows the diagnostic confidence scores obtained with CS-BH-MRCP versus RT-MRCP for both senior and junior radiologist. For the senior radiologist, diagnostic confidence was better with the CS-BH-MRCP sequence for 17 patients (25%) and with the RT-MRCP sequence for only 7 patients (10.3%). For the junior radiologist, the corresponding proportions were 21 (30.9%) and 9 (13.2%). Interobserver agreement for the assessment of the confidence diagnostic was moderate with RT-MRCP (κ = 0.59, 95% CI: 0.43-0.76) and substantial with CS-BH-MRCP (κ = 0.66, 95% CI: 0.50-0.81) (Appendix A, Table A3). Table 7. Cross-tabulation analyses: diagnostics confidence scores with CS-BH-MRCP versus RT-MRCP sequences for senior and junior readers.

Bile Duct Lithiasis, Branch Duct Intraductal Papillary and Mucinous Neoplasms and Bile Duct Anastomosis Stenosis
Based on the consensus between the two readers, CBD lithiasis was present in 10 of the 54 patients. In three patients, the calculi were visible on both MRCP sequences and on a non-MRCP sequence and confirmed by ERCP. In five patients, CBD lithiasis was visible on the RT-MRCP sequence and on a non-MRCP sequence but not on the CS-BH-MRCP sequence. Among these five patients, only one had calculi visible by ERCP and three had uninterpretable CS-BH-MRCP images (diagnostic confidence of 1). Finally, in two patients, calculi were visible on the CS-BH-MRCP sequence and on non-MRCP sequences but not on the RT-MRCP sequence. Among these two patients, only one had calculi visible by ERCP. The RT-MRCP was considered interpretable in both patients. No significant difference was found between the CS-BH-MRCT and RT-MRCP sequences for the detection of bile-duct lithiasis (p = 0.30). One stone was fortuitously discovered in one patient referred for BD-IPMN, only on CS-BH-MRCP sequence (not included in analysis). Figure 4 shows three examples of MRCP images of patients referred for suspected choledocholithiasis using RT-MRCP and CS-BH-MRCP sequences. Appendix A, Table A6 shows the number of CBD lithiasis detection based on the consensus between the two readers with each MRCP sequence and the agreement between the two sequences.
Among the 10 patients who were scanned for suspected BD-IPMN, RT-MRCP images were considered uninterpretable (diagnostic confidence of 1) for four patients, whereas no CS-BH-MRCP images were considered uninterpretable. Communication between the pancreatic cyst and the MDP was visible in five patients with RT-MRCP sequence, whereas it was visible in six patients with CS-BH-MRCP sequence. Note that, regarding the cysts other imaging features, all lesions were most likely BD-IPMN, despite the inability to systematically demonstrate a communication with the MDP. Appendix A, Table A7 shows the number of patients for whom pancreatic cyst communication was visible each MRCP sequence and the agreement between the two sequences. Among the four patients with liver-transplant recipients who were referred for suspicion of biliary anastomosis stenosis, RT-MRCP and CS-BH-MRCP images were interpretable for all patients (diagnostic confidence of 2 or 3) and none of them presented noticeable stenosis. For the diagnostic of BD-IPMN and the detection of bile-duct anastomosis stenosis in liver-transplant recipients, the sample sizes were too small to allow a meaningful statistical analysis.

Discussion
In our study, based on quantitative and qualitative evaluations, the CS-BH-MRCP sequence was preferred over the conventional RT-MRCP sequence in terms of image quality at 3T with the advantage of a much shorter acquisition time. The CBD to PBT CR was significantly better. The CS-BH-MRCP sequence demonstrated significantly better overall image quality, fewer artefacts, better background noise suppression, a better visualization of the distal CBD, of the right and left primary IHBD and the distal and proximal parts of the MPD. No significant difference was found regarding the following qualitative criteria: the visualization of the proximal CBD, cystic duct confluence, secondary IHBDs, the central part of the MPD, the entire biliary system and entire MPD.
Overall, our results are consistent with previous reports. Three studies found a significantly better overall image quality with the CS-BH-MRCP sequence at 3T [9,24,31]. Studies also showed a significantly better visualization of the CBD [9], the primary IHBD [9,24], and the cystic duct [24], with the CS-BH-MRCP sequence at 3T. Nevertheless, the literature seems to show a discrepancy concerning the visualization of the MDP. With 200 patients scanned at 3T, the study of Blaise et al. showed a significantly better visualization of the MPD [24], although no segmental analysis was performed. This is in contrast with Zhu et al.'s prospective study including 80 patients, which found a worse visualization of the MPD with a CS-BH-MRCP sequence in comparison with a conventional NT-MRCP sequence at 3T [10]. CS-BH-MRCP had thus lower diagnostic sensitivity [10]. This finding prompted the same investigators to conduct another study evaluating a modified CS-BH-MRCP sequence with a smaller field of view (FOV) and higher spatial resolution that achieved better visualization of the MPD and secondary IHBDs than the "original" CS-BH-MRCP. This modified protocol also showed higher sensitivity for detecting pancreatic duct abnormalities [11]. Another optimized CS-BH-MRCP sequence at 3T with decreased accelerator factor and a reduced FOV and matrix without changes in spatial resolution, was proposed by Song et al., and demonstrated comparable or even better image quality than conventional MRCP [32]. Overall, the MPD was also better visualized with the optimized sequence [32].
Although our study's results and previous reports suggested the superiority of the CS-BH-MRCP sequence at 3T, the same was not observed at 1.5T in some studies [24,31]. Taron et al.'s study suggested a better overall image quality with the conventional NT-MRCP at 1.5T, although the results were not significant [31]. At 1.5T, in Blaise et al.'s study, the conventional RT-MRCP acquisition showed a significant superior overall image quality with better visualization of the biliopancreatic ducts, whereas only sharpness was improved with BH-CS-MRCP [24]. In a recent study, a short single BH CS-MRCP sequence, that allowed a reduced acquisition time of 8 s, demonstrated higher scores for image quality, duct sharpness and duct visualization than the conventional NT-MRCP, a CS-NT-MRCP, and a long single BH CS-MRCP (acquisition time of 17 s) sequences, the results being not always significant for all criteria and sequence to sequence comparison [33]. This highlights the potential superiority of the CS-BH-MRCP sequence, even at 1.5T and with an even shorter acquisition time.
Unlike the proximal and distal MDP that were more clearly visualized with the CS-BH-MRCP, no difference was found concerning the central MPD. The same was observed for the secondary IHBDs. It might be partially explained by the lower number of coronal slices acquired with the CS-BH-MRCP than with the RT-MRCP (64 vs. 120 slices), resulting in a smaller acquisition volume. The central MPD and secondary IHBDs were indeed less frequently imaged with the CS-BH-MRCP, resulting in a bad duct visualization score of 1. Most of the patients were referred for suspected choledocholithiasis, the field of view was thus most likely centered on the CBD with less care being taken to cover the other pancreato-biliary ducts. Great care is, therefore, required when choosing the 3D imaging volume position most appropriate for the suspected diagnosis. The radiologic technologist must also ensure that the acquisition covers as many ducts as technically possible. A significant difference in the CBD to periductal tissues CR between the two sequences that favored the CS-CH-MRCP sequence was found in our study. Although the CBD to periductal CR values were similar to values previously reported by Seo et al., i.e., 0.92 ± 0.03 for MRCP with PI and 0.91 ± 0.03 for MRCP with PI and CS, in their study, the significant difference in CR favored the MRCP sequence without CS [23]. Note that, unlike our study, both sequences were acquired using free-breathing navigator-triggered method and the two sequences acquisition parameters differed only for the acceleration factor and repetition time [23]. In Song et al.'s study, a significantly better CBD to PTB tissues CR was achieved using an optimized BH-CS-MRCP sequence compared to conventional MRCP (0.99 ± 0.01 versus 0.94 ± 0.04, p < 0.001), with slightly higher CR values compared to our study for both sequence [32].
In our study, 3D MRCP with CS was successfully acquired during a 17 s BH in 68 patients, i.e., 97% of the original cohort of 70 patients. This result was obtained although the patients had required hospital admission and exhibited multiple comorbidities likely to cause greater difficulty with maintaining a BH compared to the general population. Our study cohort was thus representative of everyday practice. Similar success rates have been reported [10,11]. The 10 min period needed for image reconstruction precluded immediate evaluation of image quality with repeated acquisition or the performance of additional sequences if needed. This point is a limitation to the use of the BH sequence instead of the free-breathing sequence. We used the reconstruction time to prepare the next patient. However, with the recently marketed latest version of the MRI machine software, the reconstruction time is only 20 s, allowing for the fast evaluation of the image quality of the acquisition and its repetition if necessary.
In the present study, radiologist diagnostic confidence was significantly better with the CS-BH-MRCP sequence that is certainly linked to the image quality. These results tend to support the use of the CS-BH-MRCP sequence for the diagnosis of biliary and ductal pancreatic diseases. Image quality indeed needs to be good enough in order to make a diagnosis with a very high degree of certainty. However, according to the senior reader, image quality was optimal to ensure complete confidence in the diagnosis (diagnostic confidence score of 3) in only 11.8% of patients with the RT-MRCP sequence and 26.5% with the CS-BH-MRCP sequence. Moderate confidence (diagnostic confidence score of 2) was achieved in 69.1% with the RT-MRCP and in 57.4% with the CS-BH-MRCP.
ERCP was rarely performed in our study and was often delayed. Furthermore, the sometimes lengthy times between MRCP and ERCP might explain the discrepancy between their findings, since spontaneous migration of the stone to the digestive tract might occur before ERCP. In addition, for patients who underwent ERCP in other centers, the results were not always available. We were therefore unable to assess and compare the diagnostic performance of the MRCP sequences. However, our analysis did not show a significant difference for the detection of bile duct lithiasis between the two sequences. Of the 11 patients with detected lithiasis, 6 had stones visible on the CS-BH-MRCP sequence. Given the short acquisition time of this sequence, it would be of interest to determine the sensitivity of a second acquisition in the event of bad image quality on the first acquisition. Tokoro et al.'s study suggested that the addition of the CS-BH-MRCP to the conventional MRCP protocol at 3T added value to the MRCP examination, since the CS-BH-MRCP could compensate for the image deterioration of the RT-MRCP caused by motion artefacts, although the image quality of the CS-BH-MRCP was not better than the RT-MRCP [34].
We were unable to further analyze the data on pancreatic cystic lesions, due to the weak sample size of this subgroup and the absence of available ERCP results. Nonetheless, the BH sequence visualized the proximal and distal parts of the MPD more clearly compared to the free-breathing sequence. In addition, a communication between the pancreatic cyst and the MDP was slightly more often visualized with the CS-BH-sequence, allowing the diagnosis of BD-IPMN. Therefore, the BH sequence may be relevant for evaluating pancreatic duct disorders, provided the acquisition volume is well centered on the MPD. The optimized CS-BH-MRCP proposed by Song et al. showed very interesting results and significantly better demonstrated the communication between the pancreatic cyst and the MPD as compared to the conventional MRCP [32]. With 41 patients included for the evaluation BD-IPMN using MRCP at 1.5T, the short single BH CS-MRCP sequence at 1.5T proposed by Henninger et al. demonstrated significantly higher scores in all the diagnostic approach criteria (lesion conspicuity, confidence, communication) compared to the conventional NT-MRCP, a CS-NT-MRCP, and a long single BH CS-MRCP sequences [33]. CS-BH-MRCP sequences that are specifically optimized for pancreatic ducts diseases assessment, could therefore improve the diagnostic performance in this indication.
The strengths of this study included the prospective design, reading of the images by two observers, exhaustive analysis of image parameters, and large number of patients compared to the published reports. However, our study presented several limitations. First, it was a single-center study. Second, the image quality analysis was mainly subjective. Nevertheless, the analysis was performed by two readers and interobserver agreement for the assessment of image quality qualitative evaluation variables ranged from moderate to substantial. Third, the excessively small subgroup sizes of patients with pancreatic cyst and liver-transplant recipients did not provide us enough data to allow a meaningful statistical analysis. Fourth, ERCP was rarely performed to confirm the diagnosis or the results were not available. We were therefore unable to assess the diagnostic performance of the MRCP sequences. However, this study mainly focused on image quality assessment. Fifth, blinded reading of the image parameters was biased by the recognizable appearance of CS-BH-MRCP images. Sixth, many acquisition parameters differed between the two sequences, that may make comparison challenging. However, this study compared a RT sequence with a BH-CS sequence. These are fundamentally different scan procedures that cannot be performed with identical protocol parameters. As such, these differences are not a true limitation of the study design but an inherent consequence of the applied techniques.

Conclusions
The present study shows that the CS-BH-MRCP sequence provides overall better image quality and bile and pancreatic ducts visualization compared to the conventional RT-MRCP sequence at 3T, with the advantage of a much shorter acquisition time. More studies are required to determine the diagnostic performance of this sequence for pancreato-biliary pathologies.  Institutional Review Board Statement: Ethical review and approval were waived for this study, due to the routine use of all MR sequences in clinical practice.
Informed Consent Statement: Written informed consent, including consent for publication, was obtained from the patient.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to identity reasons.