Manual or semi‐automated edge detection of the maximal far wall common carotid intima–media thickness: a direct comparison

Peters SAE, den Ruijter HM, Palmer MK, Grobbee DE, Crouse JR, O’Leary DH, Evans GW, Raichlen JS, Lind L, Bots ML, on behalf of the METEOR Study Group (University Medical Center Utrecht, Utrecht, The Netherlands; Keele University, Keele, UK; Wake Forest University School of Medicine, Winston‐Salem, NC, USA; Caritas Carney Hospital, Boston, MA, USA; Wake Forest University School of Medicine, Winston‐Salem, NC, USA; AstraZeneca, Wilmington, DE, USA; and Uppsala University Hospital, Uppsala, Sweden). Manual or semi‐automated edge detection of the maximal far wall common carotid intima–media thickness: a direct comparison. J Intern Med 2012; 271 : 247–256. Background. Automated edge detection is thought to be superior to manual edge detection in quantification of the far wall common carotid intima–media thickness (CIMT), yet published evidence making a direct comparison is not available. Methods. Data were used from the METEOR study, a randomized placebo‐controlled trial among 984 individuals showing that rosuvastatin attenuated the rate of change of 2 year change in CIMT among low‐risk individuals with subclinical atherosclerosis. For this post hoc analysis, CIMT images of the far wall of the common carotid artery were evaluated using manual and semi‐automated edge detection and reproducibility, relation to cardiovascular risk factors, rates of change over time and effects of lipid‐lowering therapy were assessed. Results. Reproducibility was high for both reading methods. Direction, magnitude and statistical significance of risk factor relations were similar across methods. Rate of change in CIMT in participants assigned to placebo was 0.0066 mm per year (SE: 0.0027) for manually and 0.0072 mm per year (SE: 0.0029) for semi‐automatically read images. The effect of lipid‐lowering therapy on CIMT changes was −0.0103 mm per year (SE: 0.0032) for manual reading and −0.0111 mm per year (SE: 0.0034) for semi‐automated reading. Conclusion. Manual and semi‐automated readings of the maximal far wall of the common CIMT images both result in high reproducibility, show similar risk factor relations, rates of change and treatment effects. Hence, choices between semi‐automated and manual reading software for CIMT studies likely should be based on logistical and cost considerations rather than differences in expected data quality when the choice is made to use far wall common CIMT measurements.