Related Experiment Video
Updated: Nov 4, 2025
In vivo Structural Assessments of Ocular Disease in Rodent Models using Optical Coherence Tomography
Published on: July 24, 2020
Artificial Intelligence Efficiently Identifies Regional Differences in the Progression of Tomographic Parameters of
Purpose:
To develop an artificial intelligence (AI) model to effectively assess local versus global progression of keratoconus using multiple tomographic parameters.
Methods:
This was a retrospective review of medical records of patients diagnosed as having keratoconus. A total of 1,884 Pentacam (Oculus Optikgeräte GmbH) scans of 366 eyes (296 patients) were analyzed. Based on an increase in maximum anterior curvature (Kmax), the eyes were classified as actual "progression" and "no progression." The corresponding changes in other Pentacam parameters were incorporated to train and cross-validate (five-fold) the AI models. Three AI models were trained (an increase in Kmax by A = 0.75 diopters [D], B = 1.00 D, and C = 1.25 D). The area under the curve (AUC), sensitivity, specificity, and classification accuracy, along with other metrics, were evaluated.
Results:
The AUC, sensitivity, specificity, and classification accuracy were 0.90, 85%, 82%, and 83%, respectively, for Model A; 0.91, 86%, 82%, and 88%, respectively, for Model B; and 0.93, 89%, 81%, and 91%, respectively, for Model C. All models also predicted that 60% to 62% of the actual progression eyes had concomitant progression-associated changes in the other Pentacam parameters (global progression). However, there was discordance between increase in Kmax and concomitant associated changes in the other parameters in 38.8% to 40% of the eyes (local progression).
Conclusions:
The AI models identified the eyes where the increase in Kmax and corresponding progression-associated changes in the other parameters were in agreement. These eyes may require corneal cross-linking earlier than the rest. [.
