Files and links (1)
Free to read and download, Open Access
Coronary artery disease is a leading cause of death in the Western world. Symptoms present only late in the progression of the disease, limiting treatment options; moreover, the inability to biopsy arterial tissue in a living patient makes it difficult to study the pathology effectively. 89 patients were imaged twice at a one year interval using x-ray angiography (the traditional modality for assessment of arterial stenosis) and intravascular ultrasound (IVUS), which yields a detailed image of the structure of the vessel wall. 32 of these 89 patients were made available for analysis in this study.
The Volcano Corp.~IVUS acquisition systems include software that provides a \textit{virtual histology} (VH) characterization of plaque composition that provides information otherwise only available by biopsy. Using a geometric reconstruction method described in previous work, a full working model of wall shear stresses (WSS) produced by blood flow and vessel wall composition is created. Using these, the morphologic structural information gleaned from the 3-D reconstruction, and some additional composite indices, combined with demographic information collected at enrollment and serum biomarkers collected from each patient during imaging, a detailed portrait of each patient's disease state is created, with the objective of predicting disease evolution over a 1 year timescale.
We have, in the course of this study, accomplished the following 5 aims towards the goal of predicting localized changes in disease state on a 1 year timescale:
Aim 1: Develop and validate a method of compensating for rotational motion of the catheter within the vessel and its effect upon the 3-D orientation of the reconstruction. Aim 2: Develop and validate a method of registering the reconstructed vessels that permits identification of a point-to-point correspondence on all quantitative indices. Aim 3: Successfully reconstruct, register, and analyze image sets for each of as many patients as possible for analysis. Aim 4: Identify statistically significant indices in the data suitable for use as features in a classifier. Aim 5: Construct and assess performance of a classification system that can draw useful conclusions about the 1-year progression of the arterial pathology in a patient not used in the training set.
Aim 2 was a complete success. Branches were reliably present in the IVUS data in sufficient quantities to facilitate reliable identification of the overlap and the requisite spatial transformation required to map points from one pullback onto another.
Aim 1 was much more problematic. While a method was developed which showed promise, the image acquisition protocol did not provide for orientation of the angiograms with an eye towards bifurcation identification. With neither an analytic model, nor reasonable fiducials, the method developed could only be tested on a small subset of the data, limiting both our confidence in its validation, as well as its usability in this study. It is hoped that the method can be refined and used in any subsequent study, given proper planning during the acquisition of the images, and that in turn the spatial reliability of the reconstructions can be improved beyond what is possible today.
Regarding aim 3, 32 patients were ultimately processed completely.
Aims 4 and 5 were completed successfully. Meaningful correlations were identified in the data, and the results illustrate that while we were by no means able to obtain perfect classification, we were able to handily beat a both a random, and a maximum likelihood classifier.