Bone marrow edema pattern identification in patients with lytic bone lesions using digital subtraction angiography-like bone subtraction on large-area detector computed tomography.

OBJECTIVE: The objective of this study was to evaluate the performance of digital subtraction angiography (DSA)-like bone subtraction with 2 different registration methods for the identification of bone marrow edema pattern (BMEP) in patients with lytic bone lesions, using magnetic resonance imaging as the criterion standard. MATERIALS AND METHODS: Fifty-five patients with a lytic bone lesion were included in this prospective study with approval from the ethics committee. All patients underwent magnetic resonance imaging and low-dose computed tomographic (CT) perfusion after signing an informed consent. Two CT volumes were used for bone subtraction, which was performed with 2 different algorithms (rigid and nonrigid). Enhancement at the nonlytic bone marrow was considered as a sign of BMEP. Two readers evaluated the images blindly. The presence of BMEP on bone-subtracted CT images was evaluated subjectively and quantitatively. Image quality was assessed. Magnetic resonance imaging was used as the criterion standard. RESULTS: Using a rigid registration method, the sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of CT with DSA-like bone subtraction BMEP was 77%, 100%, 100%, 68%, and 85%, respectively. The interobserver agreement was good (κ, 0.782). Image quality was better using a nonrigid registration. With this algorithm, artifacts interfered with image interpretation in only 5% of cases. However, there was a noticeable drop in sensitivity and negative predictive value when a nonrigid algorithm was used: 56% and 52%, respectively. The interobserver agreement was average with a nonrigid subtraction algorithm. CONCLUSIONS: Computed tomography with DSA-like bone subtraction is sensitive and highly specific for the identification of BMEP associated with lytic bone lesions. Rigid registering should be preferred, but nonrigid algorithms can be used as a second option when artifacts interfere with image interpretation.