Magnetic resonance cholangiopancreatography with compressed sensing at 1.5 T: clinical application for the evaluation of branch duct IPMN of the pancreas

Magnetic resonance imaging (MRI) is the method of choice for the evaluation of cystic pancreatic lesions [1]. Among all cystic lesions, the intraductal papillary mucinous neoplasm (IPMN) of the pancreas is increasingly the focus of attention as many of these cystic lesions are discovered incidentally [2]. MRI is mainly used for its non-invasive diagnosis, especially for the branch duct–type IPMN for detecting communication with the main pancreatic duct. Magnetic resonance cholangiopancreatography (MRCP), based on heavily T2-weighted sequences, is therefore the cornerstone and an established method [3‐5]. Normally, thin-slice, isotropic, navigator-triggered (NT) 3D MRCP sequences are performed to prove ductal communication. Single-shot breath-hold 2D sequences are also frequently used, but their benefit of short acquisition time is often limited due to thicker sections and therefore the missing availability of crucial 3D reconstructions [6].

The management of IPMN is constantly evolving with currently two main strategies: the Fukuoka guidelines and the white paper of the ACR Incidental Findings Committee [2, 7]. With possible follow-up intervals of 6 months over a period of 2 years, it is important that the MRI examination is optimized for these patients. Standard and widely established MRCP protocols of the pancreas normally have long acquisition times; especially, the NT 3D MRCP sequences can take up to 7 min, strongly depending on the cooperation (regular respiratory rhythm) of the patient [8]. Furthermore, such sequences may also result in suboptimal image quality in patients who do not have a consistent breathing pattern.

Compressed sensing (CS) is a relatively recent approach for the acceleration of image acquisition and is used in different anatomic regions [9‐11]. It uses a sparse, incoherent undersampling of k-space and a nonlinear iterative reconstruction to correct for subsampling artifacts. This can preserve image quality and enables the acquisition of fast 3D MRCP datasets, even in a single breath-hold [12, 13]. Although the benefits of CS are obvious, it is important to evaluate such accelerated sequences for daily clinical routine with emphasize on robustness and diagnostic accuracy. In current literature, the use of CS for MRCP has already frequently been reported, mainly at 3 T [12‐14]. Nearly all studies found the CS approach feasible with comparable results to standard MRCP sequences, which have significantly longer examination time.

Our purpose was to evaluate MRCP with CS for the assessment of IPMN of the pancreas at 1.5 T. For this purpose, conventional navigator-triggered (NT) sampling perfection with application-optimized contrast using different flip angle evolutions (SPACE) 3D MRCP was compared with various prototype 3D CS-SPACE-MRCP sequences in a clinical setting.

Written informed consent on the MRI was obtained from each patient. Institutional review board approval was granted by means of a general waiver for studies with retrospective data analysis (local research ethics committee, Medical University of Innsbruck; 20 February 2009). The authors who are employees of Siemens Healthcare had no control of any data and were not involved in the execution of the study.

All results are summarized in Table 3.

In our study, the BHS-SPACE-CS-MRCP sequence showed the highest scores concerning image and diagnostic quality for the evaluation of BD-IPMN. Interestingly, results from BHS-SPACE-CS were noticeably better compared with the NT-SPACE-CS with significantly better results concerning all evaluated categories. The main advantage is the short acquisition time with just one very short breath-hold which also seems to influence the image quality and diagnostic confidence positively for CS-MRCP sequences at 1.5 T.

The diagnosis of the BD-IPMN is based on the proof of a communication with the main pancreatic duct. In general, this preferably is performed with thin-slice 3D sequences with longer acquisition time compared with 2D sequences [17]. We are aware that the communication with the main pancreatic duct is not always detectable with MRI in general. However, it was not the aim of our study to evaluate the diagnostic accuracy of MRCP for the diagnosis of BD-IPMN but to find the best possible and optimized MRI sequence for this purpose. In this respect, a BH-SPACE-CS sequence with a very short breath-hold seems to be an ideal sequence for the evaluation of BD-IPMN. Especially if follow-up is needed, this further reduces the required scanning time without loss of diagnostic value.

Chandarana et al compared a breath-hold accelerated SPACE sequence with sparsity-based iterative reconstruction against a conventional respiratory-triggered (RT)-SPACE-MRCP at a 3-T system [12]. The overall image quality scores for BH SPARSE-SPACE were higher than those for RT-SPACE, and it was concluded that a BH sequence showed similar or superior image quality for the pancreatic duct, despite 17-fold shorter acquisition time. This study group was one of the first to evaluate CS techniques for MRCP, and their results are comparable with ours. Most of the other so far published studies in recent literature found comparable (or even superior) results between 3D MRCP sequences performed with and without CS at 3 T; in all cases, a shortening of the examination time with constant to better image quality was reported [13, 14, 18].

Furlan et al also assessed 3D CS-MRCP for the evaluation of pancreatic cystic lesions [19]. Their study, however, had no emphasis on IPMN and was carried out on a 3-T scanner, and the evaluated CS sequence was respiratory- or navigator-triggered. The acquisition time was decreased twofold by using CS without significantly compromising the overall image quality and the degree of artifacts.

There is little literature on the evaluation of CS-MRCP sequences at 1.5 T. The study by Taron et al evaluated CS-MRCP sequences with 1.5 T and 3 T [20]. In addition to a phantom study, they evaluated 66 patients (19 for IPMN) with a conventional 3D NT-SPACE and a prototype 3D NT/BH CS-SPACE. The BH CS-SPACE resulted in significantly inferior ratings in all aspects compared with NT-CS-SPACE, and the NT-CS-SPACE was found superior to the standard NT-SPACE. These results are in contradiction to ours. We had no comparison to 3-T images, but, in our patient cohort, the CS BH sequences were superior to CS NT sequences. Further CS NT sequences had the most inferior ratings of all tested sequences. One reason for the different results could possibly be seen in the different approach of rating the sequences. In our study, the main focus was on the representation of the pancreatic duct, and we did not evaluate, e.g., the visibility of peripheral structures in the biliary system. Another reason for our good results with the CS BH sequence might also be our relatively old patient population (mean age 68 years; range, 46–89 years). Especially for these patients, the very low breath-hold time of only 7 s might be beneficial. It has been shown that short examination time in general leads to better image quality [21]. It has to be noted that in agreement with our results, other studies as well found better results with the BH CS sequences at 3 T, compared with NT CS sequences [22, 23].

Tokoro et al found that the BH CS-MRCP sequence brings added value to standard NT MRCP, because it compensates the image deterioration of NT MRCP [24]. In detail, they found 43/113 cases with clinically inadequate image quality with the NT MRCP. The image quality in 13/43 cases could be improved to clinically adequate by adding the BH CS-MRCP sequence.

Image reconstruction for the CS sequences is usually very long (approx. 5 min) [17]. By integrating the CS sequences as the last sequence into our standard protocol, the time of patient change was used for reconstruction without delaying the routine workflow. With this approach, however, there is no possibility to check image quality and to eventually rerun the sequence, as the images were only available when the next patient was already in the examination room. The time of image reconstruction was not further evaluated in our study, because our MR scanner was equipped with the standard hardware which was not optimized to reconstruct demanding compressed sensing applications.

There are limitations to this study. First, this is a retrospective study. Second, we did not perform correlation with concrete pathology; e.g., the diagnosis of BD-IPMN was not further assessed nor was there any correlation with other diagnostic procedures. Our study aimed at finding an ideal sequence and comparing the performance between several sequences. Third, we had no records of patient cooperation. This might be of interest as we hypothesized that the age could be a relevant factor for image quality which was not proven or correlated with the general condition of our patients. Finally, our study population was rather small.

MRCP imaging with CS-SPACE for the evaluation of BD-IPMN at 1.5 T provides the best results using a short breath-hold sequence with results comparable with conventional NT-SPACE and superior to NT-CS-SPACE. This approach is feasible and an excellent alternative to standard 3D NT-MRCP sequences, especially concerning the diagnosis and follow-up of BD-IPMN.