George Masganas

Image processing for semi-automatic segmentation and area determination in multi-stain histology slices from porcine coronary arteries

Master's Thesis
Date: Wed. Feb. 21, 2007

Abstract:

The goal of this project was to develop semi-automated imaging techniques to analyze the cellular/biochemical structure of coronary artery sections using histology images. In addition we investigated how to improve quantitative accuracy of results over manual techniques currently in use. Measurement of lipid content inside the artery wall defined by the inner (lumen) and outer (external elastic membrane) boundaries was selected as the case study. Inner and outer boundaries were determined using region growing and livewire techniques on Verhoeff and Oil Red O stained frozen post-mortem sections of porcine coronary arteries with a wide range of wall thicknesses, which indicated a wide range of plaque (disease) severity. Cross-stain registration was done to compare boundaries determined from Verhoeff and Oil Red O images. Lipid content was measured by counting identified pixels in the red component of Oil Red O stained sections, and compared to manual measurements. The effect of blurring, due to the tradeoff between optical magnification and field of view (FOV) in the microscope, on the measured lipid content, was investigated. Equivalent lumen and vessel wall areas obtained from the histology images were compared to measurements obtained from in-vivo intravascular ultrasound images (IVUS). Results indicate that the equivalent cross-sectional areas obtained from the histology images agree fairly well with the in-vivo IVUS measurements. It was found that using a lower magnification objective in the microscope, while desirable for increased FOV, causes significant shifts in color intensities and significant blurring at magnifications less than 10x, which reduces the quantitative accuracy even with the manual techniques currently in use. Results indicated that a color intensity calibration and/or scaling method is needed to improve the quantitative accuracy of results. This is especially important when a) lipid distribution was not homogeneous across the samples under investigation or b) the lipid was distributed as small islands. We anticipate that the approaches we have developed have potential applications for analyzing histology images obtained utilizing different stains for measuring various cellular/biochemical markers of interest.

Committee Members:
Prof. Mehmet Dokmeci (advisor)
Prof. Dana Brooks
Prof. Ahmet Umit Coskun (MIE)