Segmentation techniques of brain arteriovenous malformations for 3D visualization: a systematic review

A PRISMA flowchart is displayed in Fig. 1. A total of 7212 publications were screened, 37 full-text articles were assessed for eligibility and 32 studies were included in this review.

Fig. 1figure 1

Summary of search strategy (PRISMA flow chart) for relevant studies

Segmentation input (Table 1)

Thirteen out of 33 studies (39%) described high-resolution magnetic resonance imaging (MRI)-based segmentation [20, 27,28,29,30,31,32,33,34,35,36,37,38]. Within this group, tridimensional time of flight-MR-Angiography (TOF-MRA) was used in 11 studies (85%): 9 studies applied a protocol without gadolinium and 2 studies with gadolinium (28, 31). Nine studies (27%) described digital subtraction angiography (DSA)-based segmentations [39,40,41,42,43,44,45,46,47], and 7 (21%) described native CT and CT-Angiography (CTA)-based segmentations [25, 48,49,50,51,52,53] (Table 1).

Segmentation aim (Table 1)

Thirteen out the 33 studies (39%) aimed to render feeding arteries, the architecture of the nidus and draining veins [25, 28,29,30, 34, 37, 38, 43, 45, 47, 48, 50, 51]. Seven studies (21%) provided an analysis of feeders and veins, without specific focus on the nidus [18, 27, 31, 33, 49, 52, 54]. The exclusive focus on the nidus was documented in four studies (12%) [20, 21, 36, 54]. Information on the aim of the segmentation is provided in Table 1. Thirty studies (91%) implemented a segmentation strategy to achieve preoperative characterization of bAVMs; however, three studies (9%) segmented with the purpose to visualize on a navigation-linked intraoperative display [38, 55, 56].

Manual and semiautomatic segmentation (Table 2)Table 2 Overview on the segmentation strategies other that fuzzy-based methods

Seven studies (21%) described manual bAVM segmentation [21, 34, 35, 38, 39, 48, 55], and 6 studies (18%) described semiautomatic algorithms [27, 40, 43, 54] In this subgroup, 3 studies (23%) aimed for delineation of all the bAVM components [34, 38, 48], while the other 10 studies (77%) focused on the segmentation on a single component of the bAVM or on the volume of the lesion. Four semiautomatic segmentation studies documented a median duration of 25 (IQR 73) minutes [27, 43, 51, 54].

Automatic segmentation (Table 2)

Twenty studies (61%) used an automatic mathematical algorithm to segment bAVMs [20, 25, 28,29,30,31,32,33, 36, 37, 42, 44,45,46,47, 49, 51,52,53, 56]. Eight of these studies (40%) aimed to segment all three bAVM components [25, 28,29,30, 37, 45, 47, 51]. Median segmentation time was 10 min (IQR 33), described in 6 out of the 20 studies. Eight automatic segmentation studies (40%) performed segmentation by an unsupervised fuzzy-based method, with a median processing time of 10 min (IQR 33) [20, 28,29,30,31,32, 36]. Only 1 research group further provided a hemodynamic characterization of the segmented bAVM components. Hemodynamics were provided by integrating temporal blood flow information of the vessels in proximity of the nidus [32, 33].

Other groups included in this cohort performed an automatic image segmentation based on supervised methods [37, 49, 53]. These strategies included supervised principal component analysis [49], supervised 3D V-Net with a compound loss function [

留言 (0)

沒有登入
gif