Good interrater reliability of a new grading system in detecting traumatic bone marrow lesions in the knee by dual energy CT virtual non-calcium images

Objective: To evaluate the capacity of dual-energy computed tomography (DECT) virtual non-calcium (VNCa) images in detecting post-traumatic bone marrow lesions (BMLs) in the knee with a new grading system. Methods: DECT and magnetic resonance (MR) imaging were used to examine acute trauma of the knee in 32 patients. VNCa images were generated by dual-energy subtraction of calcium, and the lower end of the femur and upper end of the tibia each were divided into six regions for grading of bone marrow by two musculoskeletal radiologists using a four-grading system (Grade 4, very obvious lesions; Grade 3, relatively obvious lesions; Grade 2, slight or suspicious lesion on VNCa image and mild lesion on MR image; Grade 1, normal bone marrow). CT values were obtained in the BMLs. MR images were used as the reference standard. Grade 3-4 bone marrow was regarded as a positive result to evaluate the performance of VNCa images in detecting traumatic BMLs in the knee, and receiver operating characteristic (ROC) curve analysis of VNCa images for detection of knee BMLs was performed based on CT value of the bone marrow. Results: Bone marrow rating by the two radiologists showed very good consistency (kappa = 0.850 and 0.869 for VNCa and MR images, respectively). VNCa and MR images had good consistency (kappa = 0.799 for lower end of the femur; kappa = 0.659 for upper end of the tibia). When Grade 3-4 bone marrow was regarded as a positive result, the sensitivity, specificity, positive predictive value, and negative predictive value of VNCa images for detection of BMLs in the lower end of the femur were 73.5%, 98.6%, 94.7%, and 91.6%, respectively, and the values in the upper end of the tibia were 91.0%, 100.0%, 100.0%, and 95.4%, respectively. The CT values of bone marrow were (-52.5 +/- 31.3) HU in positive area and (-91.2 +/- 16.9) HU in negative area for the lower end of the femur, and those were (-51.3 +/- 30.2) HU in positive area and (-104.7 +/- 17.5) HU in negative area for the upper end of the tibia (all p values < 0.0001). The areas under the ROC curve of VNCa images for detection of BMLs were 0.875 for the lower end of the femur and 0.939 for the upper end of the tibia. Conclusion: Good interrater reliability of this new grading system in detecting traumatic BMLs in the knee by VNCa images of DECT can be obtained with good diagnostic predictive values. (C) 2015 Elsevier Ireland Ltd. All rights reserved.