Detection of occult vertebral fractures by quantitative assessment of bone marrow attenuation values at MDCT

  • OBJECTIVES: To determine a cut-off value of Hounsfield attenuation units (HU) at multidetector computed tomography (MDCT) for valid and reliable detection of bone marrow oedema (BME) related to occult vertebral fractures.

    METHODS: 36 patients underwent both MDCT and Magnetic Resonance Imaging (MRI) for evaluation of vertebral fractures of the thoracolumbar spine and were included in this retrospective study. Two readers independently assessed HU values at MDCT in a total of 196 vertebrae. Reliability was assessed by intraclass correlation coefficient and Bland-Altman analysis. For each patient we determined the vertebra with the lowest HU value and calculated the HU-difference to each other vertebral body. HU-differences were subjected to receiver operating characteristic (ROC) curve analysis to determine the diagnostic accuracy for detection of BME as determined by MRI, which served as the reference standard. Results of HU-measurements were compared with standard visual evaluation of MDCT.

    RESULTS: HU measurements demonstrated a high interrater reliability (ICC=0.984). ROC curve analysis (AUC=0.978) exhibited an ideal cut-off value of 29.6 HU for detection of BME associated with vertebral fractures with an accuracy of 97.4% as compared to 93.4% accuracy of visual evaluation. Particularly, HU-measurements increased the sensitivity for detection of vertebral fractures from 78.0% to 92.7% due to the detection of 7 of 9 occult fractures that were missed by visual evaluation alone.

    CONCLUSIONS: Assessing bone marrow density by HU measurements using the cut-off of 29.6 HU is a valid and reliable tool for detection of BME related to occult vertebral fractures in MDCT. The introduced technique may allow more accurate treatment decisions and may make further diagnostic work-up with MRI unnecessary.

  • info:eu-repo/semantics/restrictedAccess
Forschungsinformationssystem des UKE

Interne Metadaten