Improving Keratoconus Susceptibility Screening on Refractive Surgery Using Wavelet Computational Model and Learning Machine Techniques
Narrative Responses:
Purpose
To improve keratoconus susceptibility screening on refractive surgery, using corneal biomechanical data, specifically the signs of applanation and pressure derived from Ocular Responser Analyser (ORA), based on wavelet computational model and learning machine techniques.
Methods
We used the signals of applanation and pressure of 201 examinations classified as keratoconus grades I and II by Krumeich classification. Both signals were analyzed individually and jointly. For the analysis of such data it have been developed computational models from Neural Network, Decision Trees and Networks Radial Basis Function (RBF), and also Wavelets signal processing.
Results
The results were divided into two groups: (1) without the use of signal processing and (2) using signal processing. For the first group the best overall accuracy was 92.54% and the best sensitivity was 92.31%. For the second group, which used signal processing, the best overall accuracy was 93.03%, the best sensitivity was 93.85% and 99.26% was better specificity. Each of these results was found in different computational models.
Conclusion
The results demonstrate that the wavelet computational model and the learning machine techniques have the potential to improve keratoconus susceptibility screening, providing more reliable and accurate indication and results of refractive surgery.