Three-dimensional fast spin-echo (3D-FSE) sequences have been recently used to provide comprehensive knee joint assessment. While preliminary studies are encouraging (1, 2, 3), widespread use of 3D-FSE sequences in clinical practice is limited by the relatively long scan times needed to acquire images with high in-plane spatial resolution and thin slices with complete anatomic joint coverage. This study was performed to investigate the feasibility of using compressed sensing (CS) k-space undersampling to accelerate 3D-FSE imaging of the knee joint.A sagittal fat-suppressed Cube 3D-FSE sequence was performed with and without CS twice on the knees of 10 asymptomatic volunteers and once on the knees of 25 symptomatic patients using a 3T scanner (MR750, GE Healthcare, Waukesha, WI) and 8-channel phased-array extremity coil. All Cube scans were acquired using a TR/TE=1500ms/20ms, α =90o, 16cm field of view, 320 x 320 matrix, 1.0mm slice thickness, 2.0 ARC parallel acceleration in phase and slice directions, 40 echo train length, and 1 excitation. CS was performed with a 1.5 k-space undersampling factor which reduced scan time from 4:44 minutes to 3:16 minutes. For CS reconstruction, k-space data was processed with a conjugate gradient solver to give aliased images for each coil that would result from uniformly undersampled k-space data. The aliased images were then Fourier transformed to give uniformly undersampled k-space data for each coil. The k-space data were processed with regular ARC parallel imaging reconstruction and combined in a sum-of-squares to give the final magnitude image (4). SNR of cartilage, meniscus, ligament, muscle, synovial fluid, and bone marrow on Cube and Cube-CS images were measured in all asymptomatic volunteers using the addition-subtraction method (5). The Cube and Cube-CS images of all symptomatic patients were independently reviewed three times at separate sittings by two musculoskeletal radiologists. During the first and second reviews, the radiologists used the Cube and Cube-CS images with multi-planar reformats respectively to determine the presence of knee joint pathology. During the third review, the radiologists performed a side-by-side comparison of the Cube and Cube-CS images to assess the 1) the clarity of cartilage, meniscus, ligament, and muscle, 2) the presence of image artifact, and 3) the conspicuity of knee joint pathology using a 5-point scale (Cube significantly better, Cube slightly better, Cube and Cube-CS identical, Cube-CS slightly better, Cube-CS significantly better). Student t-tests were used to compare SNR between Cube and Cube-CS images. Kappa statistics were used to assess agreement between Cube and Cube-CS for detecting knee joint pathology for each radiologist and agreement between the two radiologists for detecting knee joint pathology when using Cube and Cube-CS.Cube-CS had significantly higher SNR (p=003) of meniscus and similar SNR (p=0.15-0.67) of cartilage, ligament, muscle, synovial fluid, and bone marrow when compared to Cube (Figure 1). There was near perfect to perfect agreement between Cube and Cube-CS for detecting knee joint pathology for both radiologists and near perfect to perfect agreement between radiologists for detecting knee joint pathology when using both Cube and Cube-CS (Figures 2-4). For both radiologists combined, Cube and Cube-CS were graded similar for clarity of cartilage, meniscus, ligament, and muscle in 23, 31, 32, and 12 patients respectively, while Cube was graded slightly better than Cube-CS for clarity of cartilage, meniscus, ligament, and muscle in 27, 19, 18, and 38 patients respectively. No artifacts were identified on the Cube or Cube-CS images. For both radiologists combined, Cube and Cube-CS were graded similar for conspicuity of knee joint pathology in 43 patients, while Cube was graded slightly better than Cube-CS in 7 patients which was due to better visualization of superficial cartilage lesions on the Cube images (Figure 5).Our results showed that CS k-space acceleration can provide a 30% reduction in scan time for Cube imaging of the knee joint without a corresponding decrease in SNR. The use of CS did create slightly increased image blurring which may potentially decrease the conspicuity of subtle joint pathology such as superficial cartilage lesions. However, the image blurring did not result in decreased diagnostic performance on preliminary assessment with near perfect to perfect agreement between Cube and Cube-CS for detecting knee joint pathology. Additional larger studies with surgical correlation are needed to further compare the diagnostic performance of Cube and Cube-CS for providing comprehensive knee joint assessment. Additional studies are also needed to determine the best combination of CS k-space acceleration, in-plane spatial resolution, and echo train length for Cube knee imaging to minimize image blurring while maintaining short scan times.