Dynamic MR imaging of the small bowel potentially has an important role to play in future clinical workup as it provides a complementary tool for diagnosis, monitoring of therapy and follow-up of patients with small bowel disease. However, dynamic 3D MR images of the abdomen currently can only be acquired with either a relatively low temporal resolution or a limited coverage of the intestines, since the speed of the acquisition is constrained by the lowest achievable repetition time (TR). However, full coverage of the intestines is mandatory for motility assessment and low temporal resolution imaging obscures details of motion during contraction and relaxation of the small bowel. Previous studies showed that the frequency component of different abdominal and small bowel structures ranges between 0 - 0.25 Hz¹, suggesting the temporal resolution should preferably be below 2 seconds. We hypothesize that by using compressed sensing and parallel imaging (CS-PI) reconstruction techniques, higher temporal resolution data can be acquired. In this abstract we show the feasibility of using CS-PI to obtain higher temporal resolution by acquiring and reconstructing prospectively undersampled k-space acquisitions. Additionally, a comparison with the conventional Sensitivity Encoding (SENSE) technique that is available on clinical scanners is also presented.Three healthy volunteers were scanned on a 3T Philips Ingenia MRI scanner running software release 5.1.9. A 3D balanced FFE sequence was used. The sequence parameters were: TR/TE=2.7/1.3ms, flip-angle=20^o, FOV(FH/RL/AP)=400x400x100mm and isotropic resolution of 2.5mm³, resulting in a temporal resolution of 16.4s for a fully sampled scan. Retrospective undersampling using Poisson disk trajectories was performed producing different acceleration factors: 3, 4, 5, 6 and 7. Subsequently, the CS-PI reconstruction technique, using the open-source BART toolbox², which is based on iterative self-consistent parallel imaging reconstruction (SPIRIT)³, was performed to reconstruct the undersampled data. 3D isotropic total variation penalty with 0.002 regularization was used. Additionally, a retrospective uniform undersampling was applied to mimic SENSE accelerated scans. CS-PI and SENSE reconstructions of the retrospectively undersampled data were compared to the full-sampled (gold standard) reconstruction using the structural similarity index measure (SSIM). In addition, prospective CS-PI, using Poisson disk undersampled trajectories, and SENSE accelerated scans were acquired and reconstructed with acceleration factors: 3, 4, 5, 6 and 7 producing higher temporal resolution data.

Fig.1 shows the difference between full-sampled, linear and CS-PI reconstructions of retrospectively undersampled data using a Poisson disk distribution with effective undersampling factor of 6X. The used trajectory resulted in incoherent artifacts for the linear reconstruction as shown in Fig. 1C. Using CS-PI, clean and accurate reconstruction was achieved. Fig.2 presents the difference between CS-PI and SENSE reconstructions for 6X retrospectively undersampled data with the corresponding difference maps calculated by comparing the reconstructions with the full sampled data. As shown, the SENSE reconstruction suffered from coherent artifacts and visually low-quality images. Fig. 3 shows the average SSIM values of the two reconstruction techniques with different undersampling factors for the three volunteers. In Fig.4 a prospectively acquired dataset with acceleration factors of 3, 4, 5, 6, 6.5 and 7 is shown. These acceleration factors allow for temporal resolution up to 2.3s. The scans were reconstructed using CS-PI reconstructions, resulting in very good quality images for dynamic acquisition. In prospectively SENSE accelerated scans, undersampling artifacts were apparent for acceleration factors higher than 5 as shown in Fig.5.

In this abstract we showed that higher temporal resolution imaging in the intestines could be achieved by using CS-PI. This technique outperforms the conventional approach based on parallel imaging only and therefore it can provide a valuable tool for better assessment of small bowel motility. Even acceleration up to 7 could be achieved. However, to determine the maximum achievable acceleration factor, further validation and visual scoring by radiologists is needed to assess the quality of the reconstructed data. The major drawback of the technique is the long reconstruction time, especially with a large number of receiver coil channels and a large number of dynamics. However, as shown in recent studies, coil compression techniques⁴ and GPU implementation can overcome this issue, resulting in reasonable reconstruction time.