Firat, MuratCankaya, CemCinar, AhmetTuncer, Taner2024-08-042024-08-0420220899-94571098-1098https://doi.org/10.1002/ima.22717https://hdl.handle.net/11616/100483Today, corneal refraction, height, and thickness data, which are required in the diagnosis of keratoconus, can be obtained with corneal tomography devices. Pentacam four map display presenting this data is one of the most basic options in the diagnosis of keratoconus. In this article, an artificial intelligence-based method using Pentacam images is proposed to distinguish keratoconus from healthy eyes. Axial/sagittal curvature, back elevation, front elevation, and corneal thickness map images of a total of 341 keratoconus and 341 healthy corneas obtained from Inonu University ophthalmology clinic as the data set were given as input to AlexNet, one of the deep learning models, and the feature vectors of each image were obtained and combined. The most effective features in the determination of keratoconus were determined by applying ReliefF, minimum-redundancy-maximum-relevance (mRMR) and Laplacian algorithms, which are widely used in feature extraction algorithms, to the obtained feature vector. These features are classified using the support vector machine (SVM) classifier, which has high performance in binary classification. The accuracy, specificity, and sensitivity of keratoconus detection with the proposed method were found to be 98.53%, 99.01%, and 98.06%, respectively. The developed model can support the clinician to evaluate the features of the cornea and to detect keratoconus, which is difficult through subjective assessments, especially in the subclinical and early stages of the disease.eninfo:eu-repo/semantics/openAccessdeep learningfeature selectionkeratoconusPentacam four maps refractiveArticle3251548156010.1002/ima.227172-s2.0-85124557095Q2WOS:000754260000001Q2