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

Abstract

Objectives: The 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 methods: Data 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.

Results: Mean 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.

Conclusion: This 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.