Skip to main content
. 2022 Jan 18;11(3):478. doi: 10.3390/jcm11030478

Table 4.

Characteristics of machine learning-assisted studies for detection of KC severity.

First Author Year Severity Grading
(No. of Eyes)
Definition/Classification Methods Corneal Imaging Modality Reported Sensitivity in Detection of Each Severity Level
Maeda et al. [12] 1995 Mild (15)
Moderate (18)
Advanced (11)
NA TMS-1 Mild: 100%
Moderate: 100%
Advanced: 91%
Kamiya et al. [42] 2019 Grade 1 (108)
Grad e2 (75)
Grade 3 (42)
Grade 4 (79)
Amsler–Krumeich classification AS-OCT Grade 1: 88.9%
Grade 2: 68%
Grade 3: 71.4%
Grad e4: 74.7%
Issarti et al. [33] 2019 Mild KC (220) a Self-defined Pentacam 98.81%
Issarti et al. [33] 2019 Moderate KC (229) b Self-defined Pentacam 99.91%
Bolarin et al. [35] 2020 Grade I (44)
Grade II (18)
Grade III (15)
Grade IV (15)
Grade IV plus (15)
RETICS grading Sirius Grade I: 59.1%
Grade II: 33.3%
Grade III: 40%
Grade IV: 80%
Grade IV plus: 86.7%
Velazquez-Blazquez et al. [40] 2020 Mild KC (42) RETICS grading Sirius Mild KC: 63%

a A clear cornea, tomography maps compatible with KC, a Fleischer ring at the apex base, slight thinning, and anterior and/or posterior corneal steepening; b Slit-lamp findings compatible with KC, corneal thinning at the apex, Vogt striae, a clearly visible Fleischer ring and corneal tomography compatible with KC; The severity of KC was considered to be increasing from Grade 1 to Grade 4 and for Grade I to Grade IV plus.