Relationship between 3D Morphologic Change and 2D and 3D Growth of Unruptured Intracranial Aneurysms

2D Measurements.

2D measurements of the UIAs in all scans were performed manually on the IntelliSpace Portal (Philips Healthcare) by an experienced neuroradiologist (I.C.v.d.S.). The UIA length and width were measured on the TOF-MRAs on a 0.1-mm scale using electronic calipers.^6,18 UIA length was defined as the maximum distance from the UIA neck to the UIA dome. UIA width was measured perpendicular to the measured length along the maximum width of the UIA. Individual length and width measurements were made on both the baseline and follow-up scans. 2D length and width changes were determined as the difference in the 2D length and width measurements between the follow-up and baseline scans of the same UIA of the same patient.

3D Measurements.

To make 3D quantified morphology measurements of the UIAs, we manually segmented the UIAs from the original TOF-MRAs using in-house-developed software implemented in MeVisLab (MeVis Medical Solutions). All annotations were made by drawing a contour around the UIA on axial slices of the original TOF-MRA by the neuroradiologist who made the 2D measurements. The annotation did not include the parent vessels. Annotations were first made on the baseline scan, followed by the follow-up scan of the same patient. The annotations were converted to binary masks in which voxels that were located >50% inside the contour were labeled as UIAs. The images and annotations were all resampled to the median voxel size of 0.357 × 0.357× 0.500 mm. Using a marching cubes algorithm,¹⁹ we automatically fitted a mesh to the outside of the segmented UIA. The volume and surface area of the UIA were determined on the basis of the mesh around the segmented UIA. 3D volume change was determined as the difference in volume between the follow-up and baseline scans. The size of the UIA was determined by performing principal component analysis on the voxels within the segmented UIA and calculating the major, minor, and least extent. From these values, various morphology measurements were calculated on the basis of definitions in accordance with the IBSI guidelines,¹³ including compactness 1, compactness 2, elongation, flatness, and sphericity. Compactness 1 and 2 and sphericity are different measures that all quantify how similar the morphology of the UIA is to a sphere. Elongation describes the eccentricity of the UIA by describing how long it is relative to its width. Flatness quantifies the amount the UIA is flat relative to the length. Next, on the basis of the generated 3D mesh, the mean and Gaussian curvature of the surface of the UIA was determined, allowing the principal curvatures k₁ and k₂ to be calculated. By means of these principal curvatures, it was possible to determine the shape index and curvedness (Fig 1).²⁰ Shape index and curvedness were calculated for every point on the mesh, and a median over the whole mesh of the UIA was determined. The shape index is a descriptor of the local shape of the surface of an object and is scale-invariant. The curvedness is a positive value, which describes the local curvature of the surface and is dependent on the local scale of the object. These values are rotation- and translation-invariant, and Fig 1 depicts examples of how these values vary.

• Download figure
• Open in new tab
• Download powerpoint

FIG 1.

Shape index and curvedness. Shape index and curvedness values vary in 3D shapes. The shape index is a descriptor of the local shape of the surface of an object and is scale-invariant. Shape index values range from –1 (concave “cup”) through 0 (saddle point) to 1 (convex “dome”). The curvedness is a positive value to the local curvature of the surface, which usually lies between 0 and 1 and is dependent on the local scale of the object. These values are rotation- and translation-invariant.

All measurements and segmentations were performed on anonymized data sets by a neuroradiologist (I.C.v.d.S., with 15 years of experience). 2D measurements and the 3D segmentations were performed in a different order and several months apart to prevent bias. The observer was not blinded to the time order of the scans because this reflects the clinical setting. Morphology measurements were made on both the follow-up and baseline scans. Morphologic change was considered the difference between each morphology measurement at follow-up compared with baseline.

Morphologic Changes in Relation to Continuous UIA 2D and 3D Growth.

The relation between morphologic change and UIA growth was investigated by assessing growth as a continuous 2D and 3D outcome measurement (2D size: length and width in millimeters and 3D volume in cubic millimeters). Correlations were assessed using the Pearson or Spearman correlation coefficient, in which normality was tested using a Shapiro-Wilk test. The threshold for statistical significance was P < .05.

Morphologic Changes in Growing and Nongrowing UIAs.

The whole study population was categorized into 2 groups, including either growing or nongrowing UIAs. Growing UIAs were defined according to the clinical definition of 2D growth of ≥1-mm increase in any direction between the baseline and follow-up scan.⁸ All other UIAs were categorized as nongrowing. The difference in morphology measurements between baseline and follow-up scans was determined for each UIA (morphologic change) and compared between the populations of growing and nongrowing UIAs. An unpaired Student t test was used for normally distributed data, and a Mann-Whitney U test, for non-normally distributed data, for which normality was tested using a Shapiro-Wilk test. The threshold for statistical significance was P < .05.

Morphologic Change Based on Modified z Scores.

We determined the modified z score of the morphologic change, to identify UIAs with morphologic changes that significantly differed from those in most of the study population. This allowed us to differentiate those UIAs that can be considered to change in morphology more than could be expected on the basis of the trend of morphologic change in our population. The modified z score () for each morphology measurement for each UIA was determined as , where MAD is the median absolute deviation , and x_i was each morphology measurement for each (ith) UIA. For this subanalysis, we selected the morphologic parameters that were statistically significantly related to growth (either as continuous or categoric variables) on the basis of the first analyses (parameters: flatness, shape index, and curvedness). Statistically significant morphologic change was defined as any change in morphology measurement that had a modified z score >3.5.²¹ Finally, we determined the proportion of UIAs with statistically significant morphologic change in the 2 groups of growing and nongrowing UIAs.