Fully automated endoluminal contour detection in intracoronary ultrasound images: a pre-processing for intravascular elastography

Intravascular elastography is an emerging imaging technique that maps the strain distribution. The strain within the arterial wall is caused by the pulsatile mechanical solicitation of the blood pressure during cardiac cycle. In contrary to other elastographic applications, the transducer is positioned in the lumen and not in direct contact with the tissue. Therefore, segmentation of the luminal contour is required. Numerous contour detection techniques have been developed, but their automated character is generally strongly limited by either the initialization of parameters, or the manual selection of the contour search area. The segmentation method, presented in this paper, is fully automated and still accurate. This technique is based on active contours and exploits the property that, if ultrasound image textures can be modeled by Rayleigh distributions, two different regions can be identified by two different Rayleigh parameters. The contour is thus searched as a continuous smooth closed curve, that separates optimally, in the sense of the maximum a posteriori (MAP) approach, two Rayleigh distributions, one representing the blood, the other the arterial tissue. Unfortunately the luminal border corresponds rarely to the global maximum position of the likelihood function, but rather to the position of a local maximum. Indeed the arterial wall is heterogeneous, resulting in a brightness in ultrasound images, modeled by more than one Rayleigh distribution. For these reasons, an adaptive and iterative process of contour search area reduction has been introduced, based on energy criteria. Coronary artery images from 16 patients, acquired with a 20 MHz ultrasound scanner were analyzed using the developed method. Resulting automated contours show a strong correlation with those manually drawn by two experienced observers.