Quantification of intravascular ultrasound (IVUS) images is essential in ischemic heart disease and interventional cardiology. Manual analysis is very slow and expensive. We describe an automated computerized method of analysis that requires only minimal initial input from a specialist.
This study was carried out by interventional cardiologists and biomedical engineers working in close collaboration. We developed software in which it was necessary only to identify the media-adventitia boundary in a few images taken from the whole sequence. A three-dimensional reconstruction was then generated from each sequence, from which measurements of areas and volumes could be derived automatically. In total, 2300 randomly selected images from video sequences of 11 patients were analyzed.
Results obtained using the proposed method differed only minimally from those obtained with the manual method: for vessel area measurements, the variability was 0.08 (0.07) (mean absolute error [standard deviation] normalized to the actual value; this corresponds to an error of 0.08 mm(2) per mm(2) of vessel area); for lumen area, 0.11 (0.11) (normalized), and for plaque volume, 0.5 (0.3) (normalized). Regions with severe lesions (<4 mm(2)) were correctly identified in more than 90% of cases. Specialist time needed for each reconstruction was 10 (8) minutes (vs 60  minutes for manual analysis; P< .0001).
The computerized method used dramatically reduced the time and effort needed for IVUS sequence analysis, and the automated measurements obtained were very promising.