Metrics such as the Dice coefficient or Jaccard index enable the evaluation of automatic segmentation quality. However, their values do not reflect the characteristics of the resulting image. This paper explores the differences between expert-annotated masks and those generated by the Frangi filter with a lower Dice coefficient (average Dice = 0.6) for four brain vessels structures: the basilar artery, anterior communicating artery, left middle cerebral artery and right middle cerebral artery. The analysis was conducted on vascular structures extracted from Computed Tomography Angiography images. To assess these differences, we compared segmentation results based on key morphological features, including the average distance between centerlines, average vessel thickness, average length and average tortuosity. The results showed a slight displacement of centerlines between the two segmentations, while vessel length differences were not statistically significant. However, vessel thickness and tortuosity exhibited statistically significant differences. These findings provide insight into how segmentation discrepancies affect geometric vessel parameters and may inform future improvements in automatic vascular segmentation methods.