Automated Media Segmentation in Intravascular Ultrasound Images using Geometric Principal Component Analysis

Cardiovascular diseases are some of the most common diseases with a high mortality rate. The correct diagnosis of aortic stenosis can help patients get appropriate treatment, and can decrease the hazard of death. Intravascular ultrasound (IVUS) images can be used to evaluate plaque which is intermediate between the lumen and media adventitia (MA). This paper applies morphology processing to extract areas expected to be MA regions, and represents them with an ellipse by applying the Principal Component Analysis method to locate the MA regions. We evaluated our method on the standard publicly available IVUS image dataset (more than 100 images) using three common evaluation metrics JM, HD, and PAD. The automatic segmentation results are compared with manual segmentation delineated by experts. The proposed method also was compared with other methods including Watershed, Distance Regularized Level Set Evolution, and Region Growing. The results showed that the proposed method yielded satisfactory results with JM = 0.79, HD = 12.72, and PAD = 1.51. Most of the results were better than the other methods and effective in identifying MA.