Carotid intima-medial thickness measured on multiple ultrasound frames: evaluation of a DICOM-based software system

BACKGROUND Carotid intima-media thickness (CIMT) measured by B-mode ultrasonography is a marker of atherosclerosis and is commonly used as an outcome in intervention trials. We have developed DICOM-based software that measures CIMT rapidly on multiple end-diastolic image frames. The aims of this study were to compare the performance of our new software with older bitmap-based CIMT measurement software and to determine whether a ten-fold increase in the number of measurements used to calculate mean CIMT would improve reproducibility. METHODS Two independent sonographers recorded replicate carotid scans in thirty volunteers and two blinded observers measured CIMT off-line using the new DICOM-based software and older bitmap-based software. A Bland-Altman plot was used to compare CIMT results from the two software programs and t-tests were used to compare analysis times. F-tests were used to compare the co-efficients of variation (CVs) from a standard six-frame measurement protocol with CVs from a sixty-frame measurement protocol. Ordinary least products (OLP) regression was used to test for sonographer and observer biases. RESULTS The new DICOM-based software was much faster than older bitmap-based software (average measurement time for one scan 3.4 +/- 0.6 minutes versus 8.4 +/- 1.8 minutes, p < 0.0001) but CIMT measurements were larger than those made using the alternative software (+0.02 mm, 95%CI 0.01-0.03 mm). The sixty-frame measurement protocol had worse reproducibility than the six-frame protocol (inter-observer CV 5.1% vs 3.5%, p = 0.004) and inter and intra-observer biases were more pronounced in the sixty-frame than the six-frame results. CONCLUSION While the use of DICOM-based software significantly reduced analysis time, a ten-fold increase in the number of measurements used to calculate CIMT did not improve reproducibility. In addition, we found that observer biases caused differences in mean CIMT of a magnitude commonly reported as significant in intervention trials. Our results highlight the importance of good study design with concurrent controls and the need to ensure that no observer drift occurs between baseline and follow-up measurements when CIMT is used to monitor the effect of an intervention.