Skip to main content

Table 1 Manually created errors for simulation of segmentation variations

From: An evaluation of performance measures for arterial brain vessel segmentation

Error type False-positive errors (added voxels) False-negative errors (missing voxels) Boundary precision errors (radius manipulation of segments)
Number of errors
(total = 48)
8 distinct errors with 3 intensity levels (24 errors in total) 4 distinct errors of which one has 3 intensity levels (6 errors in total) 6 distinct errors with 3 intensity levels (18 errors in total)
Name of errors Superior sagittal sinus (1,2,3) Small vessels (1,2,3) Pcom under (1,2,3)
Middle meningeal artery (1,2,3) Pcom missing Pcom over (1,2,3)
Meninges (1,2,3) ICA missing ICA under (1,2,3)
Sigmoid Sinus (1,2,3) M1 missing ICA over (1,2,3)
Orbit (1,2,3)   M1 under (1,2,3)
Skull (1,2,3)   M1 over (1,2,3)
Merge/separation (1,2,3)   
Random voxels (1,2,3)   
  1. All created errors (n = 48) in the framework are listed and divided into three groups. In parentheses, the error intensity levels for each error are specified (1:subtle 2:moderate 3:severe). Abbreviations: ICA: Internal Carotid Artery, Pcom: Posterior communicating artery M1: Sphenoidal segment of the middle cerebral artery, Random voxels: Selection of random voxels (Subtle: 1%, Moderate: 2%, Severe: 3% of all ground truth voxels) from the original TOF MRA and addition to the ground truth image. Merge/Separation: merging vessels close to each other (A2 segments or M3-M4 segments). Radius manipulation of segments are also false-positive and false-negative errors but are given as a separate category. Detailed illustrations and descriptions of errors can be found in the Additional File 1