Scientific Session 10 — Musculoskeletal - Knee and ShoulderTuesday, May 2, 2017
3082. Compressed Sensing SPACE for Fast Knee MRI: A Study of Reliability of Sensitivity for Pathology
Diaz J1, Sze T2*, Raithel E3, Goerner F4, Garcia G2 1. Johns Hopkins Hospital, Baltimore, MD; 2. University of Texas Medical Branch, Galveston, TX; 3. Siemens Healthcare GmbH, Erlangen, Germany; 4. The Queen's Medical Center, Honolulu, HI
Address correspondence to T. Sze (firstname.lastname@example.org)
Objective: In this study, we continue our work to demonstrate that compressed sensing sampling perfection with application-optimized contrast using different flip angle evolutions (SPACE), a 3D turbo spin-echo isotropic sequence that uses techniques of incoherent undersampling and nonlinear iterative reconstruction, can reduce MRI time without degradation of image quality, improve diagnostic quality of images, and increase detection of abnormalities. In this iteration of our work, we move beyond single-observer subjective interpretation of image quality to evaluate interobserver concordance as they use the new sequence compared with an existing standard protocol.
Materials and Methods: Twelve MRI examinations of the knee were performed in 11 nonconsecutive patients using a standard protocol followed by the compressed sensing SPACE (CSS) sequence. Two musculoskeletal radiologists evaluated each protocol for ligament, cartilage, and meniscal abnormalities as well as subarticular bone marrow edema. To designate the anatomic location of the abnormality, we used whole-organ MRI scoring. To designate and score the severity of abnormality, we developed our own scoring system. For every abnormality evaluated, the results from the CSS and standard protocol MRI were compared to evaluate agreement. Agreement was assessed using Cohen kappa statistics with 95% CIs provided. We also examined weighted Cohen kappa statistics with Cohen-Fleiss weights. Further, for cartilage in which the ratings were 0, 1, and 2 (for normal, partial tear or fraying, and delamination, respectively), we performed the analyses described above and dichotomized diagnoses into 0 (normal) and 1 (any degree of abnormality detected).
Results: The time of acquisition for single-knee MRI using the CSS protocol was 5 minutes 40 seconds. The time of acquisition using the standard protocol was 20 minutes 43 seconds. Perfect (100%) interobserver agreement for assessment of abnormality in the CSS protocol versus the standard protocol was noted in 58% of the articular cartilage surface subdivisions, with agreement in most of the remaining subdivisions ranging from moderate to substantial. The presence and distribution of subarticular bone marrow edema showed perfect and almost perfect agreement in 71% of the articular surface subdivisions. Medial and lateral meniscal evaluation demonstrated perfect agreement in 86% of the areas evaluated.
Conclusion: CSS can reduce MRI scan time without significant degradation of image quality and has the potential to improve the quality of images and increase the sensitivity for abnormality detection due to the ability to provide high isotropic resolution, with high interobserver reliability for detection of knee abnormalities.