Helen Li, MD
Table 2. American Telemedicine Association diabetic retinopathy telehealth validation categories.
The cost-effectiveness of telemedicine diabetic retinopathy evaluation programs has been studied.5
Critical factors include the number of diabetic patients in the target area and percentage of patients who could be evaluated with and without telemedicine.
Other studies have compared the performance of telemedicine systems for detecting diabetic retinopathy and macular edema with clinic evaluation, ETDRS seven-field stereoscopic photographs or a combination of both. These studies used different digital imaging protocols and threshold referral criteria. One paper summarized and compared some of these validation studies.6
In one study, sensitivity was 78 percent and specificity 86 percent with monochromatic, 45-degree, 640 X 480 pixel, nonmydriatic images taken of each eye to evaluate patients for ETDRS level 35 (mild nonproliferative) or higher. In another study, digital 30-degree, 1152 X 1152 pixel by 24-bit color, seven-field stereoscopic mydriatic images evaluated patients for ETDRS level 53 (severe nonproliferative) or higher with a sensitivity of 92 percent and specificity of 90 percent. Clinically significant macular edema was detected in the same study with a sensitivity of 88 percent and specificity of 94 percent through dilated pupils. A third study found a sensitivity of 27 percent and specificity of 98 percent for detecting clinically significant macular edema in diabetic eyes using a system of three-field, 45-degree, 640 X 480 X 24-bit color, stereoscopic nonmydriatic images through undilated pupils.
These results are not surprising since macular edema detection requires high image quality and resolution. My experience is that a larger retinal area is necessary for telemedicine systems designed for evaluating retinopathy severity, compared to systems used to screen for the presence or absence of disease. I have also found the faster learning curve associated with the nonmydriatic fundus cameras to be an advantage in telemedicine systems. However, even when using a nonmydriatic camera, dilating a pupil with one drop of 0.5% tropicamide is recommended if more than one image is taken per eye. Image quality degrades as pupils get smaller with additional image flashes.
In the U.S., diabetic retinopathy telemedicine systems are currently being used by some Veterans Affairs facilities,7 several Native American health services8 and some primary care providers.9 Barriers to telemedicine adoption in the U.S. include reimbursement and the need to be professionally state-licensed in both local and remote locations. Barriers are less of an issue in the United Kingdom, which has adopted digital photography for diabetic retinopathy evaluation, easing their transition to telemedicine. Two-field digital retinal photography has been an approved method since December 2006. Though not specifically about telemedicine, the English National Screening Programme for Diabetic Retinopathy's Web site provides guidelines for grading images, quality assurance, training, cameras and software.
Digital technology can be used not only for remote retinal viewing but also for computer recognition of disease signs and severity. Computer-aided detection (CAD) algorithms can already detect semiautomatically some diabetic retinopathy lesions within digital retina images.10 Computers are also being "trained" to recognize easy-to-miss but critical pathology, such as small neovascularization11 and mild venous beading.12 Other algorithms are being developed to quantify and measure lesions.13 If successful, computers could be used to consistently and objectively determine severity levels without the observational variability inherent among humans. CAD may be similarly useful in epidemiology studies or clinical trials that follow disease progression. Such systems might uncover unseen pathology in serial retinal images, providing rich opportunities to mine existing retina image databases.
The ability of computers to detect, quantify or diagnose eye disease autonomously is an emerging technology but nowhere near the point of substituting for a full, in-person eye examination. CAD system development, testing and validation is time consuming and expensive. Artificial intelligence employs inductive machine learning to extract rules and patterns out of data. Massive databases of "ground truth" eye imagery are required to initially develop algorithms for evaluating diabetic retinopathy from retina images. Algorithm performance must be subsequently validated and tested for sensitivity and specificity to achieve high true-positive rates and low false positives. Prototype CAD systems should also be battle tested in the field because computers that work well under lab conditions may not function as expected in clinical environments.
There will likely be more than technical barriers in bringing diabetic retinopathy CAD to life. The high-tech industry is known as an intellectual property minefield. Researching existing property rights, applying for new patents and defending against infringements can be prohibitively time consuming even for large organizations. Working within regulatory frameworks will be another issue. Consider radiology's experience in adopting CAD to X-rays. The FDA approved CAD for breast cancer detection in mammography in 2002 and lung nodules in chest CTs in 2003. CAD, however, has not been approved as a replacement for a radiologist. The FDA's approval is based on a partnership between a radiologist and CAD system.14
The Road Ahead
The future promises to be both exciting and tumultuous. The evolution of digital diabetic retinopathy technology may follow the same route traveled by other technological developments. The research firm, Gartner Group, describes the process of new technology development as a five-phase visibility curve or "hype cycle" that rises and falls over time (Figure 2). Initial strong interest (and sometimes alarm by conventional technology stakeholders) in a new technology creates expectations that prove hard to meet. Interest fades as research continues. Technological problems are solved, largely off the public's radar screen. Eventually, the new technology begins taking over as the new convention.
Helen Li, MD
Figure 2. The Gartner Group's "hype cycle" shows the integration of new technology as a five-phase visibility curve that rises and falls over time.
Validated tele-retinal imaging systems have been reported to increase effectiveness at bringing diabetic retinopathy evaluations into alignment with annual guidelines.9,15
As digital image technology, telecommunications and artificial intelligence play ever greater roles in diabetic retinopathy evaluation, the question of how and when physicians interact with patients becomes more interesting. Technology development cycles and future studies on clinical and economic effectiveness will ultimately determine digital technology's role in eye care. The combination of digital retinal imagery with CAD for diabetic retinopathy evaluation could significantly challenge conventional wisdom.
- Preliminary report on effects of photocoagulation therapy. The Diabetic Retinopathy Study Research Group. Am J Ophthalmol. 1976;81(4):383-396.
- Cavallerano J, Lawrence MG, Zimmer-Galler I, et al. Telehealth practice recommendations for diabetic retinopathy. Telemed J E Health. 2004;10(4):469-482.
- Grading diabetic retinopathy from stereoscopic color fundus photographs--an extension of the modified Airlie House classification. ETDRS report number 10. Early Treatment Diabetic Retinopathy Study Research Group. Ophthalmology. 1991;98(5 Suppl):786-806.
- Fundus photographic risk factors for progression of diabetic retinopathy. ETDRS report number 12. Early Treatment Diabetic Retinopathy Study Research Group. Ophthalmology. 1991;98(5 Suppl):823-833.
- Aoki N, Dunn K, Fukui T, Beck JR, Schull WJ, Li HK. Cost-effectiveness analysis of telemedicine to evaluate diabetic retinopathy in a prison population. Diabetes Care. 2004;27(5):1095-1101.
- Whited JD. Accuracy and reliability of teleophthalmology for diagnosing diabetic retinopathy and macular edema: a review of the literature. Diabetes Technol Ther. 2006;8(1):102-111.
- Conlin PR, Fisch BM, Orcutt JC, Hetrick BJ, Darkins AW. Framework for a national teleretinal imaging program to screen for diabetic retinopathy in Veterans Health Administration patients. J Rehabil Res Dev. 2006;43(6):741-748.
- Fransen SR, Leonard-Martin TC, Feuer WJ, Hildebrand PL, Inoveon Health Resesarch Group. Clinical evaluation of patients with diabetic retinopathy: accuracy of the Inoveon diabetic retinopathy-3DT system. Ophthalmology. 2002;109(3):595-601.
- Zimmer-Galler I, Zeimer R. Results of implementation of the DigiScope for diabetic retinopathy assessment in the primary care environment. Telemed J E Health. 2006;12(2):89-98.
- Abramoff MD, Niemeijer M, Suttorp-Schulten MS, Viergever MA, Russell SR, van Ginneken B. Evaluation of a system for automatic detection of diabetic retinopathy from color fundus photographs in a large population of patients with diabetes. Diabetes Care. 2008;31(2):193-198.
- Jelinek HF, Cree MJ, Leandro JJ, Soares JV. Cesar RM Jr, Luckie A. Automated segmentation of retinal blood vessels and identification of proliferative diabetic retinopathy. J Opt Soc Am A Opt Image Sci Vis. 2007;24(5):1448-1456.
- Gregson PH, Shen Z, Scott RC, Kozousek V. Automated grading of venous beading. Comput Biomed Res. 1995;28(4):291-304.
- Lee SC, Lee ET, Wang Y, Klein R, Kingsley RM, Warn A. Computer classification of nonproliferative diabetic retinopathy. Arch Ophthalmol. 2005;123(6):759-764.
- Castellino RA. Computer aided detection (CAD): an overview. Cancer Imaging. 2005;5(1):17-19.
- Conlin PR, Fisch BM, Cavallerano AA, Cavallerano JD, Bursell SE, Aiello LM. Nonmydriatic teleretinal imaging improves adherence to annual eye examinations in patients with diabetes. J Rehabil Res Dev. 2006;43(6):733-740.
Dr. Li states that she has no financial relationship with the manufacturer of any product discussed in this article or with the manufacturer of any competing product.