3D segmentation of coronary arteries based on advanced mathematical morphology techniques

Abstract

In this article, we propose an automatic algorithm for coronary artery segmentation from 3D X-ray data sequences of a cardiac cycle (3D-CT scan, 64 detectors, 10 phases). This method is based on recent mathematical morphology techniques (some of them being extended in this article). It is also guided by anatomical knowledge, using discrete geometric tools to fit on the artery shape independently from any perturbation of the data. The application of the method on a validation dataset (60 images: 20 patients in 3 phases) led to 90% correct (and automatically obtained) segmentations, the 10% remaining cases corresponding to images where the SNR was very low.

Angiographic Image Analysis

“…In [10,68], authors used such operators for 3D vessel segmentation, including brain, liver and heart vessels. [110], (c) [31], (d) [68].…”

“…In particular, deterministic atlases must not be considered as less (or more) relevant than statistical ones. 10 Note that the information on coronary arteries gathered in [26] corresponds to a sample of patients with "normalsized hearts". This example illustrates the general necessity to constraint some anatomical hypotheses if we may hope to finally obtain a useful model from a finite (and generally restricted) set of patients.…”

“…In the preliminary works on coronary modelling [26], it is mentioned that there exist three variants in the coronary trees structure: right dominant (10%), balanced anatomic distribution (80%), and left dominant (10%) 10 . In [36], such variations are considered by exhaustively modelling each induced branch distribution (in this case by integrating them into the symbolic base of knowledge).…”

Filtering and segmentation of 3D angiographic data: Advances based on mathematical morphology

“…Watersheds [78] were indeed considered for 3D vessel segmentation [58], 3D vertebrae labelling [46], 4D heart segmentation [14], or 3D brain structure segmentation from newborn brain MRI [25]. The grey-level hit-or-miss transform [48] was also considered, essentially in the field of 3D vessel segmentation [49,6]. Finally, connected filters [65] and especially those based on component-trees (described in Section 5.2) were involved in several (bio)medical applications, including 3D vessel filtering and segmentation [80,76,9], 3D brain structures segmentation [17], and 2D melanocytic nevi segmentation [47].…”

“…enables the computation of a score in x from the vesselness function defined in Equation (6). The maximal score among the scales of S is chosen for each point x as its best response ν max (x) = max s∈S {ν(x, s)}.…”