Utilização da aprendizagem de máquina e seleção de atributos para o diagnóstico de ceratocone a partir de parâmetros biomecânicos da córnea

Abstract

The present study aims to study the great gap that still exists in the early diagnosis of keratoconus, since this is a disease with important repercussions in the life of the patient. We used the biomechanical evaluation of Corvis ST, where six graphs were evaluated, and their properties submitted to the analisys of artificial intelligence, through machine learning and selection of attributes, in order to create a classifier that can help achieve the diagnosis. We used a database of 382 normal eyes and 192 eyes with keratoconus grades I or II, all evaluated by a single study collaborator (R.A.J.) at a clinic in Rio de Janeiro. We used the Support Vector Machine algorithm associated with Backward Elimination, and we obtained an area result below the ROC curve of 0.933, setting up a good classifier.