Automated characterization of plaque composition from intravascular ultrasound images

An automated method for plaque characterization from intravascular ultrasound (IVUS) images is reported. The method uses texture analysis and pattern recognition approaches to determine plaque composition after plaque regions have been semi-automatically identified in IVUS frames. Plaque texture description features include gray-level-based measures, co-occurrence matrices, run length measures, and fractal-based measures out of which the most distinguishing set of features was automatically selected and used for classification. Performance of the method was assessed by comparison to the observer-defined plaque composition. Soft plaque regions were automatically identified with 90.2% classification correctness and "hard" plaque regions with a classification correctness of 89.2%. Our results clearly demonstrate the feasibility of automated plaque classification in IVUS images.