Posters
Data integration: key to improve decision-making process in refractive surgery preoperative evaluation
Poster Details
First Author: J. Lyra BRAZIL
Co Author(s): G. Ribeiro A. Machado R. Ambrosio Jr I. Ramos
Abstract Details
Purpose:
This video presents a new software based on artificial intelligence that gathered clinical (age, flap and ablation depth) and tomographic data to improve refractive surgery screening.
Setting:
Advances in corneal tomography provided important data to aid refractive surgeons in their screening process. However, the amount of information makes parameters interpretation a major challenge.
Methods:
It was used a pre-operative exams data of 266 patients that underwent LASIK with at least 2 years stable follow-up, and 58 normal preoperative exams of patients who developed post-LASIK ectasia. It was included tomographic and refractive parameters in the construction of computational models, which were based on machine learning techniques and selection of parameters such as linear regression, neural networks and genetic algorithms. The results of sensitivity and specificity of the developed models were calculated and compared using ROC curves (Receiver Operating Characteristic Curve) and their respective areas under the curve (Area Under the Curve- AUC).
Results:
The best result was found using a logistic regression by combining 3 tomographic parameters ( IHD Point Thinnest and BAD D ) and 2 clinical parameters ( Flap thickness and depth of ablation ), reaching a sensitivity of 93.1 % , a specificity of 91.0 % and an Area Under the ROC Curve ( AUC) of 0.981 .
Conclusions:
The use of machine learning and selection techniques optimized screening for susceptibility of corneal ectasia in refractive surgery preoperative exams , allowing a safer surgical indication and more accurate results.
Financial Disclosure:
One or more of the authors receives nonNONEmonetary benefits from a company producing, developing or supplying the product or procedure presented.
