A Comparative Study of Intracranial Aneurysm Diameter Measurement Approaches

Intracranial Aneurysm are balloon like structures in the brain arteries. These are formed at weak locations in the arteries, like at vessels forks and junctions. Blood get accumulated in these balloon like structures and are risk of rupture. Intracranial aneurysm rupture lead to subarachnoid hemorrhage, wherein the blood flows in space surrounding the brain. This could be a fatal and life threatening situation. It could lead to death of the patients in couple of months or lead to permanent physical, cognitive or physiological disability. Early detection and measuring the size of the aneurysm, would be beneficial to the patient. Physicians could diagnose the severity, analyze the rupture risk and plan appropriate treatment. In the current work, we have explored 8 approaches to measure the max diameter size of the aneurysm. Dice and IoU are commonly used metrics for measuring the quality of the segmentation overlap between ground truth mask and predicted mask. But both the metrics are insensitive to boundary errors. Meaning, we might get a good dice and IoU score even when the edge alignment is poor. For a proper diameter measurement between ground truth mask and predicted mask, we have considered 3 alternative metrics, HD95(Hausdorff Distance), ASD(Average Surface Distance) and Boundary IoU. A Unet based model is used segmenting the aneurysm from 2D MIP images. The 2D MIP images are generated from TOF-MRA from the ADAM dataset. Segmentations which have Dice and IoU >70% are selected for experimentation. HD95, ASD and Boundary IoU are computed for these selected segmented masks. Based on the scores of these three boundary based metrics, the images are categorized as Excellent, Good, Moderate and Poor boundary match. We found that there were no excellent category images where the segmentation ground truth and predicted mask matched very well. About 8% of the data were good boundary match. 84% were in moderate boundary match and 8% in poor boundary match. These images are used for measuring the diameter of the predicted mask and ground truth mask.