Scientific Session 06 — Musculoskeletal - FootMonday, May 1, 2017
2985. Bone and Joint Modeling From 3D Ankle MRI: Feasibility and Comparison With Radiographs and 2D MRI
Nordeck S*, Koerper C, Adler A, Xi Y, Chhabra A. University of Texas Southwestern Medical Center, Dallas, TX
Address correspondence to S. Nordeck (email@example.com)
Objective: The purpose of this study was to evaluate feasibility of creating segmented joint models (bone surface reconstructed radiography) from 3D isotropic ankle MRI, assess various angular and joint space measurements on models and compare with corresponding radiographs and 2D MRI measurements, and determine differences in time consumption for performing the measurements.
Materials and Methods: Our study was a HIPAA-compliant series of 27 consecutive subjects with 3D ankle MRI isotropic datasets (0.65-mm voxel, fast spin-echo, proton density–weighted sequence) performed on 3-T scanners. Three subjects with incomplete radiographic series were excluded. Thus, images from 11 left and 13 right ankles in 24 subjects (16 female, eight male; mean age, 46 years; age range, 17–73 years). Segmented joint models (bone surface reconstructed radiographs) were created on all isotropic MR datasets by one observer. Three readers independently performed various blinded angular and joint space measurements on the models, corresponding 2D MRI, and radiographs at study inception and 8 weeks later. Mean and SD were calculated for each measurement. Paired t tests were performed for each pair (radiograph vs 2D MRI, radiograph vs 3D model, and 2D MRI vs 3D model) within each parameter; p values less than 0.05 were considered statistically significant. Spearman correlation coefficient between measurements was performed to assess interreader agreement. Interclass correlation coefficients (ICCs) were calculated, and time difference analysis was performed.
Results: Three-dimensional segmentation models resembling radiographs were created in all cases within 10 ± 5 minutes. Good agreement (ICC > 0.65) was noted in all reader measurements across all modalities with the exception of measuring the talocrural angle on all images, talar pitch on 3D MRI, and tibiofibular clear space on 2D MRI, which showed weak-to-poor correlation. Time consumption was significantly greater (p < 0.05) on 2D than on 3D for all measurements except lateral clear space and medial tibiotalar concavity assessments. Time consumption was noted to be significantly greater (p < 0.05) on radiography than on 3D MRI for nearly half the measurements for each reader.
Conclusion: Three-dimensional bone and joint segmentation models resembling radiographs can be constructed from isotropic MRI data. Although good correlation was noted among all readers, absolute measurement values differed. These differences are likely related to positioning variances between modalities versus the ideal positioning of the 3D reconstruction. Tighter collimation, a protocol variation, also limited evaluation in some cases because reference points normally used with radiography were not included, contributing to variance. Very good (ICC > 0.7) interreader agreement was noted for most measurements with a few measurements showing moderate to good correlation (ICC > 0.5 to 0.65). Measurements on the 3D model were significantly faster when compared with 2D MRI and generally faster than radiography. Using 3D models to obtain joint space and angular measures from MRI ankle examinations improves efficiency of objective ankle assessment. In the future, automated model generation and measurement acquisition with import into a standardized report may be feasible.