Measuring morphologic properties of the human retinal vessel system using a two-stage image processing approach

The scanning laser technique in combination with digital image analysis can be used to assess the morphology of the retinal vascular tree. Quantitative description of the retinal vascular network may provide further knowledge in pathophysiology of retinal and systemic vascular disease. Especially, for monitoring of vascular alteration in follow-up studies quantitative reproducible methods to assess the vascular morphology are essential. Therefore we developed an automatic scheme allowing the measurement of morphological properties for the use in diagnosis and therapy control. To extract the morphological properties of the retina a two-stage image analysis procedure is employed. First the image is segmented in objects using a model-based top-down, image segmentation scheme. Then the obtained objects are classified with a neural net, the result being the tree of the arteries and veins. The third step is a measurement process which yields the desired information of arterial diameter, tortuosity and other morphological properties. As an example we show a functional image of the diameters and present a pilot-study in patients with arterial hypertension to demonstrate the ability of the new method for computerized analysis of the retinal vascular tree to detect arteriolar vascular alterations