Elaboration of a semi-automated algorithm for brain arteriovenous malformation segmentation: initial results

ObjectivesThe purpose of our study was to distinguish the different components of a brain arteriovenous malformation (bAVM) on 3D rotational angiography (3D-RA) using a semi-automated segmentation algorithm.Materials and methodsData from 3D-RA of 15 patients (8 males, 7 females; 14 supratentorial bAVMs, 1 infratentorial) were used to test the algorithm. Segmentation was performed in two steps: (1) nidus segmentation from propagation (vertical then horizontal) of tagging on the reference slice (i.e., the slice on which the nidus had the biggest surface); (2) contiguity propagation (based on density and variance) from tagging of arteries and veins distant from the nidus. Segmentation quality was evaluated by comparison with six frame/s DSA by two independent reviewers. Analysis of supraselective microcatheterisation was performed to dispel discrepancy.ResultsMean duration for bAVM segmentation was 64 ± 26 min. Quality of segmentation was evaluated as good or fair in 93 % of cases. Segmentation had better results than six frame/s DSA for the depiction of a focal ectasia on the main draining vein and for the evaluation of the venous drainage pattern.ConclusionThis segmentation algorithm is a promising tool that may help improve the understanding of bAVM angio-architecture, especially the venous drainage.Key points• The segmentation algorithm allows for the distinction of the AVM’s components• This algorithm helps to see the venous drainage of bAVMs more precisely• This algorithm may help to reduce the treatment-related complication rate