950-93 Automatic Identification of Calcified Regions on Intravascular Ultrasound Images of Human Coronary Arteries Using Texture Analysis

The potential for intravascular ultrasound (IVUS) to evaluate arterial pathologies and to suggest invasive techniques thus far has been limited by its qualitative and subjective evaluation. The purpose of this study is to develop tools to automatically identify and objectively quantify the calcifications present in IVUS images of human coronaries.Human coronary arteries were dissected from explanted hearts within one hour of transplantation and perfused with saline at physiological pressure.IVUS images were acquired using an HP Sonos 100 system and a 3.5 Fmonorail 30 MHZ catheter. A controlled pullback was recorded on SVHS videotape and images digitized every 250 μm. Texture analysis, consisting of the convolution with several different Laws’ matrices followed by texture “energy” extraction and nearest neighbor classification, was used to segment each image into high- and low-information regions. The regions of low-information content corresponded to shadows behind echo-dense features such as calcium. Calcifications were identified as sectors of radially thin texture regions that correlate with bright areas in the intensity image function. Angular percentage of vessel cross section with calcifications and angular localization of calcifications were compared with manual tracing in a set of 230 images from patients with different pathologies. The comparison showed a good correlation: Percentage correct: 86%, Sensitivity: 79%, Specificity: 89%. No preferential angular calcification localization were observed.ConelusionThe results show that this is an objective method to clearly detect and quantify calcifications in IVUS images