View abstract

This abstract is assigned to session AOB Free Papers 1

Presentation type	Oral presentation
Title	Training the “Cone Location and Magnitude Index Expanded” on Keratoconus Suspect Eyes using Machine Learning

Purpose	To use artificial intelligence in retraining the Cone Location and Magnitude Index Expanded to detect keratoconus suspect corneas.
Methods	349 eyes of 349 patients from the American University of Beirut Medical Center were included in the study, with 133 eyes with normal control corneas (NL), 147 eyes with definite keratoconus (KC), and 75 eyes with keratoconus suspect corneas (KCS). KCS corneas were normal fellow eyes of patients with highly asymmetric keratoconus. Different parameters from anterior, posterior corneal map and pachymetry were used.
Results	Using logistic regression, the machine learning model trained on three categories of eyes achieved a sensitivity of 67%, specificity of 87%, positive predictive value (PPV) of 58%, and F1 score of 0.62 in detecting KCS. A new artificial intelligence algorithm (CatBoost) scored 71% of sensitivity, 91% of specificity, a PPV of 68% and F1 score of 0.70, while using the same parameters incorporated in the original index.
Conclusion	Artificial intelligence improves the performance of the Cone Location and Magnitude Index Expanded in detecting early suspicion of keratoconus.
Conflict of interest	No
Details of conflicting interests	None

Authors 1

Last name	WEHBI
Initials	Z
Department	American University of Beirut
City	Beirut, Lebanon

Authors 2

Last name	Hammoud
Initials	B
Department	American University of Beirut
City	Beirut, Lebanon

Authors 3

Last name	Assaf
Initials	J.F
Department	Casey Eye Institute
City	Portland, Oregon

Authors 4

Last name	Awwad
Initials	S.T
Department	American University of Beirut
City	Beirut, Lebanon