Lumen detection in human IVUS images using region-growing

To assess the health of arterial tissue in intravascular ultrasound (IVUS) images, detection of luminal borders is critical. The authors have enhanced two automated border detection schemes using region-growing based on inter-pixel gray-scale differences, components-labeling (CL) and dilation-erosion, and watershed segmentation (WS) to correct for leaked regions due to signal drop-out and strut artifacts. The two methods were evaluated using 8 IVUS images with and without calcium lesions. Shapes were quantitatively analyzed, and the cross-sectional lumen areas calculated from the two automated methods were compared with the areas from expert traced images. Algorithm execution times were also compared. Results: CL vs. expert traced had a mean area difference of 7 pixels (p>0.05), WS vs. expert traced had a mean area difference of 394 pixels (p<0.05), for the leaked images. Thus region-growing with CL accurately predicts luminal areas of the artery and corrects for luminal leaks better than WS.