Intracranial aneurysm measurements: Inter-observer variability for two measuring techniques

I. LARRABIDE; J. M. MACHO; L. SAN ROMAN; J. BLASCO; E. E. VIVAS; D. RUFENACHT; M.C. VILLA-URIOL; R. CARDENES; J. M. POZO; D. R. HOSE; A. F. FRANGI

Houston

Conferencia; Live Interventional Neuroradiology Conference (LINC) in conjunction with Intracranial Stent Meeting (ICS) 2010; 2010

Texas Heart Institute

Purpose Geometrical descriptors of intracranial aneurysms are largely used for diagnosis and treatment selection nowadays. Nevertheless, relatively little work has been dedicated to automate the computation of these measurements and to eliminate inter-observer variability. Measurements are manually obtained by radiologists and/or clinicians with the subsequent subjectivity and variability upon different observers and dependency on the view point. In this study we investigate the inter observer variability for two measuring techniques, one manual (method 1) and one semi-automatic (method 2).Materials & methods A total of 30 intracranial aneurysms from the @neurIST database were considered in this study. For Method 1, three independent clinical observers measured three variables of interest (namely neck width, sac height and sac width) on the volume rendering of the 3DRA images. For Method 2, the vasculature geometry of the aneurysms was manually isolated by 3 independent observers (experts in image quantification) and the three variables automatically calculated using a computer program. A two sample t-test was performed, assuming equal variances and normal distribution of the aneurysm measurements. Results Each of the three variables was studied. The null hypothesis indicated that observations corresponded to populations with different variance. For all the measurements, namely neck width (p=0.004), sac height (p=0.20) and sac width (p=0.007), a low significance was observed between Method 1 and Method 2. Regarding repeatability, the standard deviation (and 95% repeatability coefficient, which indicates the maximum difference likely to occur between two measurements under the assumption of no bias) was computed for both methods using the three observations. For Method 1 vs. Method 2, inter-observer variability for neck width (0.47 (1.32) vs. 0.17 (0.49)), sac height (0.75 (2.08) vs. 0.12 (0.35)) and sac width (0.51 (1.42) vs. 0.53 (1.46)) were found to be larger (3 to 5 fold) in the case of Method 1 always but for the sac width. Conclusions Poor agreement was observed between clinical measurements and semi-automated measurements. Such discrepancies might be attributed to the limitations in the volume rendering visualization as well as to the differences in the measurement criteria among observers. The existence of automated measuring tools that are properly validated should effectively reduce these differences.