David S. Shin

Computer-assisted, interactive fundus image processing for macular drusen quantitation

PURPOSE To design and validate a software package to quantitate the area subtended by drusen in color fundus photographs for the conduct of efficient, accurate clinical trials in age-related macular degeneration. DESIGN Algorithm and software development. Comparisons with manual methodologies. PARTICIPANTS Evaluation and testing on color fundus photographs from patient records and from eyes enrolled in the Choroidal Neovascularization Prevention Trial. METHODS Fundus photographs of eyes with drusen were digitized. The green channel was selected for maximum contrast and preprocessed with filtering and shade correction to minimize noise, improve contrast, and correct for illumination and background inhomogeneities. Local thresholding and region-growing algorithms identified drusen. Multiple levels of supervision were incorporated to maximize robustness, accuracy, and validity. Validation studies compared computer-assisted with manual grading by an experienced grader. Intraclass correlation coefficients were calculated as a measure of the concordance between manual and computer-assisted fundus gradings. MAIN OUTCOME MEASURES Drusen area and concordance with manual grading. RESULTS Automated supervised image analysis offers extreme robustness and accuracy. Most images were segmented with little or no supervision, with processing times on the order of 5 seconds. More complicated images required supervision and a total analysis time varying from 20 seconds to 5 minutes, with most of this time devoted to inspection and comparison. Interactive image processing affords arbitrarily close concordance with manual drusen identification, with calculated intraclass correlation coefficients of 0.92 and 0.93 for comparison of manual with automated, supervised grading by two observers. CONCLUSIONS Automated supervised fundus image analysis is an efficient, robust, valid technique for drusen quantitation from color fundus photographs. This approach should prove useful in the conduct of efficient accurate clinical trials for age-related macular degeneration.

Laser-induced drusen reduction improves visual function at 1 year

OBJECTIVE To describe the relationship of laser-induced drusen reduction to change in visual function at 1 year among patients enrolled in the Choroidal Neovascularization Prevention Trial (CNVPT). DESIGN Comparison of groups with and without drusen reduction; follow-up of a randomized controlled trial. PARTICIPANTS Evaluations of drusen and visual acuity at baseline and at 1 year were performed for 351 eyes of the 432 eyes enrolled in the CNVPT Bilateral Drusen Study and Fellow Eye Study (81%). One hundred eighty-four eyes were assigned to observation, and 167 eyes were assigned to laser treatment. Eyes with conditions that precluded an analysis of drusen reduction, such as those that developed choroidal neovascularization (CNV) within the first year, are excluded from this analysis. METHODS Change in macular drusen between initial visit and after 1 year was assessed by side-by-side grading by evaluators masked to information on visual function. Visual acuity, contrast threshold, and critical print size were measured by certified visual function examiners. MAIN OUTCOME MEASURES Change in visual acuity is the primary outcome. Change in contrast threshold and change in critical print size are secondary outcome measures. RESULTS Laser-treated eyes with 50% or more drusen reduction at 1 year had more 1- and 2-line increases in visual acuity and less losses in visual acuity compared with laser-treated eyes with less drusen reduction or with observed eyes (P = 0.001). Similar improvements were noted for contrast threshold but not critical print size at 1 year. CONCLUSIONS Laser-induced drusen reduction is associated with improved visual acuity and contrast sensitivity in eyes at 1 year. Longer term effects of laser-induced drusen reduction on visual function require additional observation. The overall potential value of laser treatment in eyes with high-risk drusen requires consideration of not only short-term effects on vision but also the effects of CNV and atrophy on vision.

Computer-vision-enabled augmented reality fundus biomicroscopy

PURPOSE To guide treatment for macular diseases and to facilitate real-time image measurement and comparison, investigations were initiated to permit overlay of previously stored photographic and angiographic images directly onto the real-time slit-lamp biomicroscopic fundus image. DESIGN Experimental study in model eyes, and preliminary observations in human subjects. METHODS A modified, binocular video slit lamp interfaced to a personal computer and framegrabber allows for image acquisition and rendering of stored images overlaid onto the real-time slit-lamp biomicroscopic fundus image. Development proceeds with rendering on a computer monitor, while construction is completed on a miniature display interfaced directly with one of the slit-lamp oculars. Registration and tracking are performed with in-house-developed software. MAIN OUTCOME MEASURES Tracking speed and accuracy, ergonomic acceptability. RESULTS Computer-vision algorithms permit robust montaging, tracking, registration, and rendering of previously stored photographic and angiographic images onto the real-time slit-lamp fundus biomicroscopic image. In model eyes and in preliminary studies in a human eye, optimized registration permits near-video-rate image overlay with updates at 3 to 10 Hz and misregistration errors on the order of 1 to 5 pixels. CONCLUSIONS A prototype for ophthalmic augmented reality (image overlay) is presented. The current hardware/software implementation allows for robust performance.

Computerized stereochronoscopy and alternation flicker to detect optic nerve head contour change

PURPOSE/BACKGROUND Stereochronoscopy, a technique previously explored but abandoned for glaucoma diagnosis, viewed optic nerve images acquired at separate points in time as if a stereo pair. Prior efforts to exploit this technique were impaired by a lack of superimposability for sequential optic nerve images. We investigated computerized registration techniques for aligning sequential, monoscopic optic disc images to facilitate sensitive detection of optic nerve head contour changes in glaucoma. DESIGN Algorithm and software development. Comparisons with standard techniques. MATERIALS Existing patient records from the Glaucoma Service, Scheie Eye Institute, University of Pennsylvania. METHODS Two sets of optic disc photographs, separated in time by 1 to 18 years, of 25 eyes with and without glaucomatous optic disc progression were digitized. We developed custom software for accurate image alignment. Change in disc morphology was then judged by digital stereochronoscopy and user-controlled alternation flicker of superimposed, time-separated images on a computer monitor. Comparisons were made with standard stereoscopic comparison. MAIN OUTCOME MEASURE Identification of change or no change in optic nerve head contour for images acquired at separate points in time. RESULTS Image processing and registration permits accurate alignment of optic disc photographs. Alternation flicker of superimposed, sequential images facilitates image comparison and detection of change as indicated by change in vessel position, color, and other cues for contour change. A high concordance was found between standard stereoscopic comparison and alternation flicker. In several cases, reinspection of stereo comparison led to a revised judgment on the basis of disc changes rendered more obvious with alternation flicker. Digital stereochronoscopy was less concordant with standard techniques. CONCLUSIONS Digital image processing techniques and alternation flicker provide a simple, sensitive, software-based method for detecting glaucomatous optic disc change.

Grading, image analysis, and stereopsis of digitally compressed fundus images.

Purpose: To investigate the effects of image digitization and compression on the ability to identify and quantify features in color fundus photographs. Methods: Color fundus photographs were digitized as tagged image file format (TIFF) and high‐compression (80:1) and low‐compression (30:1) joint photographic experts group (JPEG) images. Rerendered images were subjected to standard grading protocols developed for a clinical trial, and digitized images were subjected to image analysis software for drusen identification and quantitation. Re‐created stereoscopic images were compared subjectively with originals. Results: Original, TIFF, and low‐compression (30:1) JPEG images were virtually indistinguishable when subjected to close scrutiny with magnification. The overall quality of high‐compression (80:1) JPEG images and images digitized at 500 dots per inch was markedly reduced. Protocol grading of original and digitized images was highly concordant within the repeatability of multiple grading of original images. The area subtended by drusen differed by less than 1.0% for all uncompressed and compressed image pairs quantified. Stereoscopic information was accurately preserved when compared with originals for TIFF and low‐compression JPEG images. Conclusions: Fundus images can be digitized and stored with significant compression while preserving stereopsis and image quality suitable for quantitative image analysis and semiquantitative grading. Low‐compression (30:1) JPEG images may be suitable for archiving and telemedical applications.

Fundus image change analysis: geometric and radiometric normalization

Image change analysis will potentiate fundus feature quantitation in natural history and intervention studies for major blinding diseases such as age-related macular degeneration and diabetic retinopathy. Geometric and radiometric normalization of fundus images acquired at two points in time are required for accurate change detection, but existing methods are unsatisfactory for change analysis. We have developed and explored algorithms for correction of image misalignment (geometric) and inter- and intra-image brightness variation (radiometric) in order to facilitate highly accurate change detection. Thirty-five millimeter color fundus photographs were digitized at 500 to 1000 dpi. Custom-developed registration algorithms correcting for translation only; translation and rotation; translation, rotation, and scale; and polynomial based image-warping algorithms allowed for exploration of registration accuracy required for change detection. Registration accuracy beyond that offered by rigid body transformation is required for accurate change detection. Radiometric correction required shade-correction and normalization of inter-image statistical parameters. Precise geometric and radiometric normalization allows for highly accurate change detection. To our knowledge, these results are the first demonstration of the combination of geometric and radiometric normalization offering sufficient accuracy to allow for accurate fundus image change detection potentiating longitudinal study of retinal disease.

Image-guided macular laser therapy: design considerations and progress toward implementation

Laser therapy is currently the only treatment of proven benefit for exudative age related macular degeneration and diabetic retinopathy. To guide treatment for macular diseases, investigations were initiated to permit overlay of previously-stored angiographic images and image sequences superimposed onto the real-time biomicroscopic fundus image. Prior to treatment, a set of partially overlapping fundus images is acquired and montaged in order to provide a map for subsequent tracking operations. A binocular slit-lamp biomicroscope interfaced to a CCD camera, framegrabber board, and PC permits acquisition and rendering of retinal images. Computer-vision algorithms facilitate robust tracking, registration, and near-video-rate image overlay of previously-stored retinal photographic and angiographic images onto the real-time fundus image. Laser treatment is guided in this augmented reality environment where the borders of the treatment target--for example, the boundaries of a choroidal neovascularization complex--are easily identified through overlay of angiographic information superimposed on, and registered with, the real-time fundus image. During periods of misregistration as judged by the amplitude of the tracking similarity metric, laser function is disabled, affording additional safety. Image-guided macular laser therapy should facilitate accurate targeting of treatable lesions and less unintentional retinal injury when compared with standard techniques.