Retinal Imaging Background

Still digital images and live digital image sequences of the human retina are captured using a digital video camera attached to a fundus camera [42, 85]. Still photography is generally performed with a flash and live imaging is done with continuous illumination. In the former case, imaging noise is lower than in the latter case where less light is available. The confocal imaging case is not considered here [116]. Readers who do not have access to a source of retinal images, such as an ophthalmology clinic, can access large collections of retinal images on CD-ROM [123].

Several aspects of retinal images make automated processing challenging. First, the images are highly variable. Large variability is observed between images from different patients, even if healthy, with the situation worsening when pathologies exist. For the same patient, variability is observed under differing imaging conditions and during the course of treatment. Unlike industrial vision problems where the conditions can be carefully controlled, retinal images are frequently subject to improper illumination, glare, fadeout, and loss of focus. Artifacts can arise from reflection, refraction, and dispersion.

The movement of the eye can result in significant variation, even for a stable, well-seated patient. The eye movements are rapid and significant [88]. The saccadic movements involve sudden j umps up to 15°. These movements occur at speeds ranging from 90 to 180°/sec. The mean peak acceleration can range from 15,000 to 22,000°/sec2. These movements cannot be fully suppressed, even with medication. Often, medications are considered harmful, and so are not used. These numbers imply that interlaced image sensors are often a poor choice for imaging the vasculature [4]. Most commercial imaging systems, such as the TOPCON ImageNet system, employ non-interlaced digital megapixel cameras. Additionally, movements resulting from a distracted patient, or in response to the irritation induced by the laser, or the involuntary attempt to fixate the fovea on the laser light, all affect the quality of the image. Finally, each image or frame in a video sequence represents a partial view of the retina, and the zoom factor (magnification, scale) can vary from frame to frame, due to either selection of a different magnification setting on the fundus camera or movement of the camera nearer to or farther from the patient's eye (which may be necessary for focusing).

The inherent difficulty of illuminating the retina in a uniform and steady manner for reflectance imaging [100] causes several difficulties. For effective imaging, the retina cannot simply be illuminated by directing a strong light into the patient's eye. In addition to causing patient discomfort, this would generate reflections off the patient's cornea (glare) that can dominate the light reflected from the retina back out the pupil. As a result, another method of retinal illumination is used. Before imaging a patient's retina, the patient is administered eye dilating drops (e.g., Tropicamide ophthalmic solution) to dilate the pupil to allow a large area for light to enter and exit the eye. The fundus camera's illumination is focused to form a ring of light at the anterior (front) of the eye with an unilluminated circle in the center of the pupil. As a result, reflection off the surface of the eye only occurs in the illuminated ring. At the retina, the light ring is out of focus, and the result is diffuse illumination. The aperture of the fundus camera is designed to block the reflected light from the circle of illumination while allowing light from within the unilluminated circle, resulting in a fundus image uncontaminated by reflection or glare from the cornea. Unfortunately, this method does not always work flawlessly. For the retina to be illuminated while preventing glare, the camera must be focused properly on the patient's eye. Movements of the patient's eye, even slight ones, can degrade the focusing. In particular, if the eye moves so that the iris blocks some of the ring of illumination, the total illumination of the retina can drop dramatically. In addition, eye movements can allow some stray reflected or refracted light to be directed into the aperture of the fundus camera. This is seen as a bright glare at the edges of the retinal image. Figure 6.6 shows examples of these possibilities.

When live video imaging is desired, one has to contend with eye movements and the difficulty of providing steady illumination. The factors noted above result in image frames that are often dim, out of focus, motion blurred, or corrupted with optical effects such as glare or nonuniform illumination. In retinal still photography, skilled technicians quickly refocus the camera for optimum illumination for each picture. Images are only taken/saved when the illumination is optimal. Any images with insufficient illumination or excessive glare are simply discarded. Tracing algorithms designed for real-time

0 0

Post a comment