Abstract
Purpose :
Image and pattern recognition is key to clinical competence in ophthalmology. Internet Image Search, notably Google Image Search (GIS) in the U.S., has become an important teaching tool for ophthalmology trainees and allied health staff. Users utilize search results to enhance their own foundry of reference images for each disease. However, Google results are presented by a search-rank algorithm which does not take into account the accuracy of the image to what was searched. We quantify the accuracy of results and assess the quality of the image source in several fundus pathologies.
Methods :
Ten common and 10 uncommon fundus pathologies were randomly selected from an ophthalmic text. GIS was performed and the top 10 consecutive results were analyzed. Results were graded as having adequate (> 300 pixel resolution) or poor image quality, correct diagnosis, peer-reviewed source, brief, comprehensive or no supporting text, and correct supporting text. Correctness of diagnosis was evaluated by review of the image and context in the supporting text. The image source was classified as originating from professional ophthalmic education, professional medical education, general health information, ophthalmic health information, private practice, ophthalmic retail, medical association, peer-reviewed journal, book or private websites.
Results :
In total, 200 images and source websites were evaluated in the study. The majority were of adequate quality (74%). Most images were fundus photographs (80%) and a minority contained multiple image modalities (30%). Overall, 7.5% were of the incorrect diagnosis. Among common pathologies, 1% were of the incorrect diagnosis while 14% of uncommon pathologies were incorrect. Half (51%) of the top image results were from professional ophthalmic education websites. 30% of images were from peer-reviewed sources. About one third (30%) of websites had only brief descriptions of the pathology while 20% had no supporting text. 10% of images were published on private websites created by non-experts. Ten websites contributed 46% of all resulted images. Images of the incorrect diagnosis were presented most often due to mis-attribution to part of the search phrase (e.g. drusen in “dominant drusen”).
Conclusions :
Internet Image Search rapidly generates a collection ophthalmic pathology for study. Attention should be paid to the accuracy of images originating from a myriad of sources.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.