Near real-time plaque segmentation of IVUS

This article is devoted to the creation of a near-real time framework to discriminate between tissue and blood. We perform a fast supervised learning of local texture patterns of the plaque using local binary patterns. A classifier is built by assembling weak classifiers using boosting schemes that allow quick performance and reliability. After that, a deformable model is used to ensure continuity in the segmentation and to fill in the gaps in the classification scheme. Our supervised learning framework has been validated using 450 test images from 15 different patients. The resulting segmentation differs from the physicians segmentation in a mean rate of 0.15 mm. and maximum rate of 0.33 mm. The method benefits from the low time consuming feature extraction, as well as a faster classification scheme reducing 10 times the whole processing time compared to most of the texture based approaches.