Task 1: Detection
The algorithm output should be a 3D coordinate (x,y,z) of the voxel at the binary centre of mass of the candidate aneurysm (s). Each candidate location should be on a new line in the text file. The coordinates of each candidate location should be in the same image coordinates as the original TOF MRA. Positive detection will be determined based on the predicted candidate location coordinates (x,y,z) being located within the maximum aneurysm radius based on the manual aneurysm mask.
Detection algorithm output: a text file “result.txt” under the /output folder containing only the 3D voxel coordinates of the detected aneurysms. Each detected aneurysm should be given on a different line of the text file in format x, y , z
Multiple detections are allowed and likely to occur, and each candidate location should be provided on a separate line. No other punctuation or text should be in the text file other than two commas between the coordinates.
Example text file contents for the detection of two aneurysms:
233, 132, 80
241, 60, 12Note: Do not provide an aneurysm radius for the detection task, only a centre of mass coordinate. The radius used to assess positive detections will be determined by the organisers based on the ground truth aneurysm binary mask.
Task 2: Segmentation
The algorithm output should be a binary mask of the predicted segmented aneurysm in the same image space as the original TOF MRA.
Segmentation Algorithm Output: a binary image, where the predicted unruptured aneurysm(s) is/are labelled with value 1, and background has label 0. The image should be saved in compressed nifti format, as
“result.nii.gz” under the /output folder.
For multiple aneurysms, a separate aneurysm will considered one that is a 3D connected component. Also, please note that methods submitted for participation in task 2, will also be assessed for task 1. In this case, the candidate location of each aneurysm will be determined as the centre of the mass of the aneurysms provided in the binary masks.
For both tasks
Detection/segmentation of the treated (e.g. coiled) aneurysms will not be considered when assessing the performance of untreated, unruptured aneurysm detection or segmentation. That is: any false positive detections at the location of treated aneurysms will be ignored during evaluation.
As the test set is secret, the evaluation of the results will be performed at the organisers’ institute. The aneurysm mask, centre of mass and radius are provided for each training case so the participant can self-evaluate based on this. The source code that will be used for the final evaluation can be found here.