Print Get Citation Citation Disclaimer: These citations have been automatically generated based on the information we have and it may not be 100% accurate. Please consult the latest official manual style if you have any questions regarding the format accuracy. AMA Citation Shah H, Shroff Karhade D. Shah H, & Shroff Karhade D Shah, Harsh, and Deepti Shroff Karhade. Artificial-intelligence, deep-learning algorithm identifies papilledema from fundus photographs. 2 Minute Medicine, 11 May 2020. McGraw-Hill, 2020. AccessNeurology. https://neurology.mhmedical.com/updatesContent.aspx?gbosid=549911§ionid=246698533APA Citation Shah H, Shroff Karhade D. Shah H, & Shroff Karhade D Shah, Harsh, and Deepti Shroff Karhade. (2020). Artificial-intelligence, deep-learning algorithm identifies papilledema from fundus photographs. (2020). 2 minute medicine. McGraw-Hill. https://neurology.mhmedical.com/updatesContent.aspx?gbosid=549911§ionid=246698533.MLA Citation Shah H, Shroff Karhade D. Shah H, & Shroff Karhade D Shah, Harsh, and Deepti Shroff Karhade. "Artificial-intelligence, deep-learning algorithm identifies papilledema from fundus photographs." 2 Minute Medicine McGraw-Hill, 2020, https://neurology.mhmedical.com/updatesContent.aspx?gbosid=549911§ionid=246698533. Download citation file: RIS (Zotero) EndNote BibTex Medlars ProCite RefWorks Reference Manager Mendeley © Copyright Clip Full Chapter Figures Only Tables Only Videos Only Supplementary Content Top Artificial-intelligence, deep-learning algorithm identifies papilledema from fundus photographs by Harsh Shah, Deepti Shroff Karhade Listen +Originally published by 2 Minute Medicine® (view original article). Reused on AccessMedicine with permission. +1. An artificial-intelligence, deep-learning algorithm was shown to differentiate between the diagnosis of papilledema and normal optic nerve from ocular fundus photographs. +2. The negative predictive value was shown to be high for the developed deep-learning system to identify papilledema from ocular fundus photographs. +Evidence Rating Level: 1 (Excellent) Study Rundown: + +Optic nerve examination is a fundamental component to detect papilledema. However, the examination through direct ophthalmoscopy is avoided or poorly performed by nonophthalmic specialists. Currently, artificial intelligence and deep learning are automatically detecting ocular pathologies such as glaucomatous optic neuropathy from ocular fundus photographs. As such, this study trained, validated, and externally tested a deep-learning system to identify and classify normal optic disks and disks with papilledema. The system was trained and validated using 14,341 fundus photographs. Furthermore, the system was externally tested on 1,505 fundus photographs. The study found a trained and validated algorithm with high specificity and sensitivity to differentiated between papilledema and normal optic nerves. +This retrospective study was limited by the method of acquiring fundus photographs after pharmacologic pupil dilation. As such, the algorithm was calibrated to artificially dilated pupil, which may not reflect general practice. Therefore, the algorithm would not be appropriate for fundus photographs without pupil dilation. Another limitation was the retrospective study design, which caused an unequal sample collection of various optic-disk conditions and convenience sampling. In regard to the algorithm, the output would be biased based upon the skewed collection of the input samples. Nonetheless, this study was strengthened by the multiethnic population, which allowed the algorithm to adjust for physiological differences of the optic nerve amongst different ethnic groups. For physicians, these findings provide a functional algorithm to utilize for the diagnosis of papilledema when an ophthalmic specialist is not present at the practice. +Click to read the study in NEJM +Relevant Reading: Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs In-Depth [retrospective cohort]: + +This diagnostic study retrospectively collected 15,846 fundus photographs from 6,779 patients at 24 centers in 15 countries including the United States, Thailand, and France. Photographs were obtained from digital fundus cameras after the eye had undergone pharmacologic pupillary dilation. The images were centered around the macula or optic disk; however, the images always included the optic disk. Two different datasets were created to develop the deep-learning system. The first dataset consisted of 14,341 photographs from 19 sites, and the dataset was used for training and validation of the algorithm. The second data set consisted of 1,505 photographs from five other sites, and the dataset was used for external testing. Inclusion criteria included: fundus photographs of optics disks and definite corresponding clinical diagnoses made by neuro-ophthalmologists. Exclusion criteria included: presence of more than one ocular pathology on photograph and insufficient diagnostic uncertainty. Neuro-ophthalmologists provided a specific diagnosis for each photograph, and each patient had been seen by a neuro-ophthalmologist for evaluation. The deep-learning classification model consisted of a segmentation network (U-Net) to detect the optic disk location and a classification network (DenseNet) to classify the optic disk. A five-fold cross validation was performed on the first (validation) dataset, after which, the same thresholds were used to assess the classification model on the independent external dataset. The primary outcome was area under the receiver-operating-characteristic curve (AUC), which evaluated the performance of the algorithm to classify the optic disk appearance. In the validation dataset, the system discriminated disks with papilledema and all other optic disks with an AUC of 0.99 (95% confidence interval [CI], 0.98-0.99), a sensitivity of 93.2% (95% CI, 91.8-94.5), and specificity of 95.1% (95% CI, 94.7-95.6). In the external dataset, the system detected disks with papilledema with an AUC of 0.96 (95% CI, 0.95-0.97), a sensitivity of 96.4% (95% CI, 93.9-98.3), and a specificity of 84.7% (95% CI, 82.3-87.1). Additionally, with a mean prevalence of papilledema of 9.5% in the external dataset, the positive predictive value for the system was 39.8% (95% CI, 36.6-43.2), and the negative predictive value for the system was 99.6% (95% CI, 99.2-99.7). Taken together, the study concluded an artificial-intelligence, deep-learning algorithm was trained and validated to identify papilledema from ocular fundus photograph with high specificity and sensitivity. +©2020 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.