F. Knezevich

An Image Based Auto-Focusing Algorithm forDigital Fundus Photography

In fundus photography, the task of fine focusing the image is demanding and lack of focus is quite often the cause of suboptimal photographs. The introduction of digital cameras has provided an opportunity to automate the task of focusing. We have developed a software algorithm capable of identifying best focus. The auto-focus (AF) method is based on an algorithm we developed to assess the sharpness of an image. The AF algorithm was tested in the prototype of a semi-automated nonmydriatic fundus camera designed to screen in the primary care environment for major eye diseases. A series of images was acquired in volunteers while focusing the camera on the fundus. The image with the best focus was determined by the AF algorithm and compared to the assessment of two masked readers. A set of fundus images was obtained in 26 eyes of 20 normal subjects and 42 eyes of 28 glaucoma patients. The 95% limits of agreement between the readers and the AF algorithm were -2.56 to 2.93 and -3.7 to 3.84 diopter and the bias was 0.09 and 0.71 diopter, for the two readers respectively. On average, the readers agreed with the AF algorithm on the best correction within less than 3/4 diopter. The intraobserver repeatability was 0.94 and 1.87 diopter, for the two readers respectively, indicating that the limit of agreement with the AF algorithm was determined predominantly by the repeatability of each reader. An auto-focus algorithm for digital fundus photography can identify the best focus reliably and objectively. It may improve the quality of fundus images by easing the task of the photographer.

A computer algorithm to quantitatively assess quality of digital optic disc images.

Optic disc photography is used in the management and study of glaucoma. Quality assessment is needed at the time of acquisition and during review. A computerized algorithm for objective quality assessment was developed to mimic the procedure used by human observers. It was tested on film-based images obtained with mydriasis (40 normal and 46 glaucomatous eyes) and non-mydriatic digital images (30 normal and 38 glaucomatous eyes). The image sharpness was graded by six masked readers into four categories. The area under the receiver operating characteristic curve for identifying unreadable images was 1.0 for the digital and film-based images and 0.91 and 1.0 for differentiating between unreadable and mediocre images for digital and film-based images, respectively. This pilot study demonstrates that the algorithm can identify all unreadable images. Further studies are necessary to test whether it can be applied to images obtained in other locations on the fundus and with additional cameras.

A machine vision method for automated alignment of fundus imaging systems.

BACKGROUND AND OBJECTIVE To report a method to track the pupil in three axes simultaneously prior to imaging the fundus. PATIENTS AND METHODS The system is based on parallax optical alignment to detect the center of the pupil. The system consists of two cameras acquiring pupil images from two distinct directions and an operator-supervised algorithm to derive the coordinates of the pupil center and output of commands to drive a three-axes computer-controlled stage. The system was tested in a cohort of 45 individuals 61 ± 15 years of age, 26 with and 19 without glaucoma. The tracking was performed without pharmacologic pupil dilation. RESULTS The variability of the pupil center determination (assessed by the standard deviation) was ± 0.19 and ± 0.33 mm for the vertical and horizontal directions, respectively. The processing time of the algorithm was 0.75 msec. The tracking converged to within a preset tolerance of ± 0.5 mm in 45 of the 45 eyes. CONCLUSION Image acquisition and processing of pupil images can be used to align fundus imaging systems rapidly, accurately, and with minimal operator intervention.

Retinal thickness in eyes of older normal individuals and its implication for the diagnosis of glaucoma.

PurposeRetinal thickness (RT) is a useful measurement for describing diseases that affect the thickness of the retina, such as glaucoma. Existing normative data are derived from relatively young individuals; however, glaucoma is most prevalent in older individuals. We therefore studied the RT in older normal individuals. Patients and MethodsParticipants of the Baltimore Eye Study, persons accompanying patients, and staff were recruited and underwent visual field testing and a comprehensive eye examination by a glaucoma specialist. RT was measured with the retinal thickness analyzer (RTA, Talia Technology) and RT values in specific regions were derived using a custom-designed MatLab program. ResultsOne hundred and three eyes of sixty-two individuals were studied. Mean age was 61 years. Sixty-six percent were female and 82% were of European descent. The average mean deviation on visual field testing was 0.03 dB and the average pattern SD was 1.51 dB. The mean RT of the entire macula was 159±16 μm, and was lowest in the foveal pit and highest in the parafoveal annulus. The average distance from the foveal pit to the thickest point in the parafoveal annulus was 1240±138 μm. The mean RT of the entire macula was slightly less in older individuals (slope=−5.7 μm/10 y, P=0.02) but the height of the parafoveal annulus relative to the foveal pit, which is determined by the combined thickness of the parafoveal nerve fibers, ganglion cells, inner plexiform layer, and inner nuclear layer, did not vary with age (P=0.62). ConclusionsAlthough the average RT of the entire macula was slightly thinner with increasing age, the height of the parafoveal annulus relative to the foveal pit did not change with age and would therefore seem to be a better marker of neuronal tissue health than the average RT of the entire macula.