ABSTRACT
Purpose:
To evaluate the ability of horizontal thickness profile to distinguish keratoconus from normal corneas, and compare the accuracy of these indices with more complex tomographic indices.
Methods:
In a retrospective study, one eye randomly selected from 225 patients with bilateral keratoconus and 335 patients with normal corneas were included. All patients were examined by a corneal specialist and underwent the examination of corneal tomography, Pentacam HR (Oculus, Wetzlar, Germany). Data of horizontal pachymetric profile passing through the apex of 6 mm was evaluated. With these data, the thickness at the corneal apex (P.ápice), the thickness of the thinnest point of the horizontal profile (P.min.H), the pachymetric progression from the thinnest point to the thickest (PPmin-max), the average pachymetric progression in the meridian as reference the average from the normal population (PPméd) and the fitness with a second degree polynomial trendline (R2). An artificial intelligence model was built to combine this data. The performance for distinguishing normal keratoconus was evaluated by ROC curves.
Results:
All of these indices were statistically different between the two groups (p <0.001). The area under the ROC curve (AUC) thickness at the apex and TP were 0.904 and 0.938, respectively. The tomographic index with higher AUC was BAD-D (0.997). Regarding the horizontal profile, the AUC of PP.min.H and PPmin-max were 0.915 and 0.927, respectively. The fitness to the trendline to the horizontal thickness graph AUC was 0.896. The best performance was obtained with the PPméd (AUC 0.932 sensitivity = 84.4% and specificity of 92.5%). The artificial intelligence model combining all items derived from horizontal profile improved performance (AUC 0.991 Sensitivity = 96% and specificity of 98%).
Conclusion:
The horizontal thickness profile can detect keratoconus, with a capacity comparable to more complex indices. This type of analysis can provide the basics for new approaches, using data present in simpler devices than the tomographer reducing the cost for the patients.
Keywords:
Keratoconus/diagnosis; Coneral diseases; Córnea/anatomy & histology; Tomography/standards
