Purpose: To develop an application (TopEye) in the iOS platform for mobile devices to allow the capture and interpretation of color maps generated by corneal topographers using artificial intelligence.
Methods: In the execution, follow-up, and assessment of the project, we used the Scrum methodology and interactive and incremental development process for the project management and agile software development. The ge nerated diagnostic pattern bank consists of 1,172 examples of corneal topography, divided into 275 spherical, 302 symmetrical, 295 asymmetrical, and 300 irregular patterns (keratoconus). For the development of the artificial intelligence of the application, network training was established with 240 images of each pattern type, with a total of 960 patterns (81.91%). The remaining 212 images (18.09%) were used to test the application and will be used for the results. The process is semi-automatic, so the topographic image is captured with a smartphone, the examiner performs the contour of the corneal relief manually, and then the neural network performs the diagnosis.
Results: The application diagnosed 201 cases (94.81%) correctly. In 212 images, the algorithm missed the classification of 11 cases (5.19%). The major error that occurred was in distinguishing between symmetrical and asymmetrical classes. In keratoconus screening, the application reached 95.00% sensitivity and 98.68% specificity.
Conclusion: The work resulted in obtaining an efficient application to capture topographic images using a smartphone camera and their interpretations through applied artificial intelligence.