Automated Characterization  of Elastic Lamina (EL) Structures*

Modeling of measurement of biological structures of blood vessel tissues is a basic procedure to research in  the growth and remodeling (G&R) processes of the   cardiovascular system under stress conditions, e.g.,  hypertension. The biological structures of  the dissected blood vessel tissues taken from experimental subjects can be measured and characterized after they are imaged by a digital  microscope.  Understanding of these structural changes is essential to evaluation of disease conditions, and effectiveness of their medical treatments. Major challenges in characterization of the biological structures of the blood vessel tissue include (1) lack of standard reference coordinate system, (2) noisy images, and (3) labor intensive and error prone procedures. Most studies chose a sample area to assert  the EL thickness. The image noise can be caused by the cutting procedure or non-optimal selection of the chemical dye.  As a result, most study reported EL thickness results after the data are aggregated together. Although this practice maybe adequate for long term study, it may not applicable  for the early development stage of hypertension, see the examples in figures 1 (a) and (b).   For its complexity,  many  other biological structures were not considered in the current study. 

(a)

(b)

Figure 1: a) Normotensive aorta and b) 4-week Hypertensive aorta

Our lab is developing automated structural analysis tools [3] to extract these parameters for automated characterization of
the hypertensive aorta tissues collected from a large scale live animal (pigs from the same gene pool) based study by our collaborator*. In [3], we already designed a system which is able to extract some interesting geometric parameters from the aorta samples, such as the thickness of blood vessel, lamina thickness (black lines in figure 1). Key features of the fully automated EL analysis tool are summarized as follows.

• Measured parameters:   EL thickness, interlamellar distance, furcation points of EL, EL lengths. Some major geometric parameters of a zoomed tissue area  are illustrated in figure 2.
• Denoising: Artifacts produced from tissue cutting can be removed  based on the thresholds set by the user.
• Flexible sampling: Align sample tissue orientation and then make cross-section measurements along aorta wall.
•  Principal direction: A Radon transform (RT) based  algorithm is robust to noise and accurately detect the principal direction (PD) of aorta samples.
•  High level statistics: 
   

  Figure 2: Illustration of some of aorta geometric parameters

Experiment Result:

To classify the input images into 2 classes: normotension and hypertension, we tested the LDA, PCA, Snapshot PCA, Feature Subset Selection (FSS), Quadratic Classifier and Linear Classifier. Based on the experiment outcomes, LDA with 15 selected features was used to generate the feature space. A linear classifier was used rather than conventional quadratic classifier with Gaussian kernel.  Based on above models, our system is able to provide a 83.0% accuracy detection rate over 129 testing sample images, figure 3.

Accuracy = 92.3% (a)

Accuracy = 83.0% (b)

Figure 3: Aorta Sample Image Classification Results of a) Training Data, b) Testing Data

  Reference:

1. T. W. Fossum, W. I. Baltzer, M. W. Miller, M. Aguirre, D. Whitlock, P. Solter, L. A. Makarski, M. M. McDonald, M. Y. An, and J. D. Humphrey, "A novel aortic coarctation model for studying hypertension in the pig," Journal of Investigative Surgery, vol. 16, pp. 35-44, Jan-Feb 2003.
2. R. L. Gleason and J. D. Humphrey, "A mixture model of arterial growth and remodeling in hypertension: Altered muscle tone and tissue turnover," Journal of Vascular Research, vol. 41, pp. 352-363, 2004.
3. H. Xu, J.-J. Hu , J. D. Humphrey, and J.C. Liu, "Modeling and Measurement of Elastic Laminae in Arteries," 2006, pp. 626-629.

*A collaborative research effort with  Dr.   J. D. Humphrey of the Department of Biomedical Engineering, Texas A&M University