Table 1

The summary of the previous works related to keratoconus diagnosis

AuthorsDatasetClassesMethodAccuracy(%)Objective
Al-Timemy et al, 202114534 cases/4 maps (Iraq)2EDTL with AlexNet and product fusion98.30Improved accuracy
Al-Timemy et al, 2021154844 corneal images
7 maps		2EfficientNet-b097.70High accuracy in identifying KCN, time-efficient framework, low complexity.
384.40
Kuo et al, 202016354
Images/1map		2VGG16,
InceptionV3
ResNet152		93.10
93.10
95.80		High accuracy CNN
Lavric et al, 2019173000 image/1map2CNN99.33Reducing diagnostic errors and facilitating treatment.
Kamiya et al, 201918543 image/6maps2ResNet1899.10Diagnostic accuracy of keratoconus
Al-Timemy et al, 2023191371 eyes
(3 maps) (Egypt)		3Xception and InceptionResNetV297.00–100Detection of clinical and subclinical forms of KCN
213 eyes (Iraq)2Xception and InceptionResNetV288.00–92.00
Xie et al, 202020images6465/4maps5InceptionResNetV294.70Performance of the AI classification system
Aatila et al,2020212924 image (5 maps)3VGG16, InceptionV3, MobileNet, DenseNet201, Xception and EfficientNetB0.99.31and 98.51 by
DenseNet201 and VGG16		Improve performance of keratoconus classification
Otuna-Hernández et al, 202322950 image with augmentation
(corneal profile dioptre pachymetry ART)		4CNNsSevere:
Sen. Of 92.59 Spec. of 98.68		Early detection keratoconus
Fassbind et al, 2023231940 cornea scans2CNNs95.45Predicting the most common corneal diseases
593.52
Elsawy et al,202024413 eyes (4maps)4AlxN,
VGG16
VGG19		99.12
99.96
99.93		Early detection of corneal diseases

  • ART, Ambrosio relational thickness; CNN, convolutional neural network.