Yuli's Assignment 14

Title: Coronary Structure Identification in the Continuous Images

Traditional coronary angiography is performed on two-dimensional (2-D) projection views. Compare to the 2-D views, a three-dimensional (3-D) view which is reconstructed by two view geometry method could provide much more information in surgery and intro-coronary device analysis [1]. To reconstruct a 3-D coronary tree, identifying the coronary structure in advance is necessary and it’s doing manually in practice.

The objective of the research is finding an algorithm to automatically identify the structure of human arterial tree. The interested data are a set of continuous images of the interested arterial tree. In this dissertation, an algorithm is provided in 4 steps: (1) Extracting the coronary features and the structure hierarchy in each frame. The features I extract are the terminals and intersections of the vessels. By drawing the line between every two terminals and the characteristic of vessels, we can find the structure and distinguish the main vessel from side branches. (2) For each feature, identify its place in all the images (from diastole to systole) (3) Apply the machine learning (neural network) to modify the hierarchy and structure and minimize overall error in the sequenced images. (4) Compare the results with the expert opinions and further modify the algorithm if it’s necessary.

Compare to the human data, the animal data are easier to obtain and have simpler structure. Therefore, the near goal is to automatically identify the animal coronary structure. In the end, I hope this algorithm could analyze the human data in practice.

[1] S.-Y.J. Chen, J.D. Carroll, J.C. Messenger, "Quantitative analysis of reconstructed 3-D coronary arterial tree and intracoronary devices", Medical Imaging, IEEE Transactions on Volume 21, Issue 7, Jul 2002 P.724 - p.740

Last modified 9 December 2007 at 11:33 pm by Yuli