A centerline-based estimator of vessel bifurcations in angiography images

Bifurcation detection in 3D vascular images using novel features and random forest

“…It is shown that our proposed features outperform those in [20] by a large margin in crossvalidation. Both numerical and visual results show that the proposed bifurcation detector gives superior result to those obtained by the method of Macedo et al [14].…”

“…The second column shows results of VESSEL12 dataset, although the two curves almost overlap at the top, the red is much steeper than the blue at the bottom, which shows a lower tendency of detecting false positives than the competing algorithm. Figure 3 shows a visual comparison between tracking results using two bifurcation detectors, green rings indicate errors in bifurcation detection of the competing method of Macedo et al [14]. Macedo et al [14].…”

“…In our second experiment, we evaluate our method in comparison with a second competing method [14], by replacing the bifurcation detection step in Algorithm 1 (Line 8). The two new tracking algorithms are tested on 10 volumes from VascuSynth dataset (each containing 16 bifurcation points) with 4 different levels of noise (Gaussian, std = 0, 5, 10, 15) and 10 sub-volumes (50 × 50 × 50) selected from the VESSEL12 dataset.…”

“…Our method is evaluated on two datasets, VascuSynth [10] (synthetic, 3D) and VESSEL12 (clinical CT chest images, 3D), against two competing algorithms, [20] and [14]. In all the following experiments, we fix these parameters by emperical tests: α = 0.5, β = 0.5, c = 5 (c = 50 for VESSEL12 dataset), smin = 1, smax = 3 (see [9]) and C = 6, k = 11, T h v = 0.5, T h I = 50.…”

“…Note that the bifurcation detection step (Line 8) can be implemented via our or a competing method. The daughter branch detection (Line 10) is implemented using method in [14]. Centralization step (Line 16) is performed using the method in [6].…”