OCR Text |
Show COLLEGE OF ENGINEERING UNDERGRADUATE RESEARCH ABSTRACTS IDENTIFICATION OF CHOROIDAL NEOVASCULARIZATION USING OPTICAL COHERENCE TOMOGRAPHY Jared Spendlove (Mary Elizabeth Hartnett, Brittany Coats) Department of Bioengineering, Moran Eye Center, Department of Mechanical Engineering University of Utah Background Neovascular age-related macular degeneration (AMD) is the number one cause of central vision loss and blindness and affects an estimated 14 million people worldwide. Even though millions of people are affected by A M D , little is known about the mechanisms that cause the disease. To investigate signaling pathways for choroidal neovascularization (CNV), an important symptom of A M D , w e currently reproduce C NV in an in vivo transgenic mouse model. Lesions are created in the retinal layers neighboring the choroid and resolution is tracked overtime using confocal microscopy. One of the future goals of the research is to replace confocal microscopy with optical coherence tomography (OCT) imaging to monitor the same animal over time. This would result in stronger statistical correlations and less cost to animal life. The objective of this study was to create a MATLAB program that will read in a series of O C T images and calculate the width, area, and volume of the lesions created in the mouse model. Methods The quality of the O C T images was first enhanced using image filtering techniques and contrast corrective algorithms. Images were then converted into binary images to segment the retina and lesion structures. The location of the lesion was identified based on peak values and slope characteristics of the binary histogram. Once the lesion was segmented from the retina, lesion width, area, and volume were calculated. Program accuracy was determined by comparing results to human calculations of lesion dimensions from OCT images. It was expected that volumetric data collected from O C T would differ from confocal microscopy, so trends in lesion resolution were compared instead. Results The program was able to identify the presence of lesions in 76.7% of the images. The volumes calculated by the program were on average 3 5 % smaller than the volumes calculated by hand. W h e n comparing week-to- week lesion volume trends calculated from the program to volume calculated manually from confocal microscopy, the algorithm displayed trends that did not follow the trends of confocal microscopy. When comparing lesion width calculated by the algorithm to the hand calculations, the widths were comparable to manual calculations from the same O C T images within a 5%error. Conclusion The difference in volume calculations can be attributed to the algorithm not locating 1 0 0 % of the lesion segments in the O C T images. The segments of lesion that were not identified contained a significant portion of the overall lesion volume. The size and shape of the lesions played part in whether the algorithm could accurately locate the lesions. Improvement in O C T resolution by image averaging or light attenuation correction is likely to decrease error and improve accuracy in O C T evaluation. Jared Spendlove Brittany Coats 25 |