Purchase this article with an account.
Saad Y Al-Kadhi, Loai Skouti, April Ellis, Qun Zeng, Tarun Sharma, Stephen H Tsang, Tongalp H Tezel; THE USE OF ARTIFICIAL INTELLIGENCE FOR THE DETECTION OF STARGARDT’S DISEASE FROM FUNDUS IMAGES. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1501. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Developing an artificial intelligence model to diagnose Stargardt’s disease from color fundus photos.
Color fundus photos of eyes with established diagnosis of Stargardt’s disease (n=25) and normal fundi (n=25) were used to train a Convolutional Neural Network (CNN) containing three convolutional layers followed by two fully connected layers ending with a sigmoid activation function (TensorFlow, https://www.tensorflow.org/). The numeric output the networks’ diagnostic ability was expressed as a percentage value with the smallest corresponding to 0% and the highest to 100%. Values >50% were considered diagnostic for Stargardt’s disease. Sensitivity and specificity of the network’s ability to correctly diagnose Stargardt’s disease was tested against an experienced retina specialist’s clinical assessment.
The model was tested with another set of fundus photos that was not used for the training purposes. The model was able to correctly diagnose fundus photos with Stargardt’s disease (n=5), (values ranging 60.5-93.9%) vs. normal fundi (n=10), (values ranging 15.7-35.3%), (100% specificity and 100% sensitivity).
The deep learning model was able to diagnose correctly fundus images of Stargardt’s disease with 100% specificity and 100% sensitivity.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
This PDF is available to Subscribers Only