Details
- Supervisors
- Faculty
- Degree label
- Abstract
- We propose methods to analyze lipidic domains in red blood cells from fluorescent images. Confocal images and wide-field images are considered. We use prior information about cells and domains geometry to segment cells and detect lipidic domains. Methods for cell segmentation are marker controlled watershed algorithms. Markers are different depending on the used microscope and are either based on Circular Hough Transform results or on thresholding. To detect lipidic domains, two different approaches are proposed. The first method is based on iterative thresholding. The second one consists in filtering the image with a Mexican Hat filter to increase image intensities at domains locations. Then, the image is thresholded with a unique threshold defined as the one that maximizes the number of regions that correspond to domains. Results are assessed by comparison to a ground truth provided by biologists. The developed methods are used to analyze domains properties : the density of domains, their location and their area. Properties between healthy red blood cells and elliptic ones (from patients suffering from Hereditary Elliptocytosis) are also compared and methods results regarding this comparison are also assessed.