Inter-observer Reliability of CT Attenuation Measurement of Lumbar Vertebral Bodies

Abstract

Purpose/Aims

To assess reliability of lumbar vertebral body computed tomography (CT) attenuation measurement between different observers.

Rationale/Background

The International Society for Clinical Densitometry (ISCD) guidelines for DXA interpretation include assessment of “opportunistic CT” as a surrogate for DXA scan using L1 vertebral body attenuation, with threshold >150 and < 100 Hounsfield units (HU) estimating the likelihood of normal bone density and osteoporosis, respectively. ISCD guidelines include precision analysis of DXA, but there are no formal guidelines for assessing precision error when assessing bone mineral density (BMD) by CT attenuation of lumbar vertebral body. Measurement of precision have been published and we sought to determine inter-rater reliability and to assess precision by test-retest of the same patient.

Methods

Utilizing Visage PACS to view CT images, six observers each measured CT attenuation of L1 and L5 vertebral bodies of the same set of 31 separate CT scans. Measurements were performed as previously described.3 Average HU within an elliptical region of interest (ROI) of the L1 and L5 vertebral bodies were recorded for each measurement, as well as L1 and L5 ROI area. Intra-class correlation (ICC) was calculated for each of these variables, with >0.9 indicating excellent agreement, 0.75-0.9 indicating good agreement, 0.5-0.75 indicating moderate agreement, and < 0.5 indicating poor agreement. ICC was calculated of L1 attenuation measured by a single observer on a separate set of 12 patients with CT scans done within 30 days of each other. Additionally we calculated root mean square–coefficient of variation (RMS-CV) of L1 vertebral body attenuation on this set of 12 patients.

Results

ICC of L1 attenuation and L5 attenuation were 0.94 and 0.92, respectively, indicating excellent agreement between observers. ICC of ROI areas at L1 and L5 ROI were 0.04 and 0.03, respectively, indicating poor agreement (Table 1). ICC of L1 CT attenuation on repeat scans within 30 days by a single observer was 0.97, indicating excellent agreement between two readings . Root mean square-SD was 14.6 HU. Least significant change was 40.4 HU. Percent coefficient of variation was 34.6.

Implications

This study demonstrates that measurement of CT attenuation at L1 and L5 between different observers is reliable while area of region of interest at L1 and L5 between observers showed poor agreement. In test-retest of scans performed within 30 days on the same patient, a short time period in which little change is expected, measurement of CT attenuation also showed excellent agreement.