Table 1.
Characteristics of the included studies.
| 1st author/Year | Country | KC Stages | Total Samples/Cases | AI model | Diagnosis Method/parameters | Validation |
|---|---|---|---|---|---|---|
| ∗Cao et al.19 (2020) | Australia | Early KC | 88/49 | MM | Pentacam/11 | Internal |
| ∗Ahn et al.16 (2022) | South Korea | Early KC/KC to control | 69/38 | FcNN, XGBoost | Pentacam/4 | Internal & External |
| ∗Shi et al.17 (2020) | China | Early KC/KC to control | 121/33 | NN | UHR-OCT & Pentacam HR/49 | Internal |
| ∗Kuo et al.18 (2020) | Taiwan | Early KC/KC to control | 354/28 | CNN | TMS-4 | Internal |
| Issarti et al.9 (2019) | Belgium | Early KC | 389/77 | FNN | Pentacam/28 | Internal & External |
| Cao et al.20 (2021) | Australia | Early KC | 267/145 | RF | Pentacam/810 | Internal & External |
| Castro-Luna et al.23 (2020) | Spain | KC to control | 60/30 | NB | CSO/16 | Internal |
| Tan et al.22 (2022) | China | KC to control | 354/143 | FNN | Corvis ST/4 | External |
| Kamiya et al.21 (2019) | Japan | KC to control | 543/304 | CNN | AS-OCT, CASIA/6 | Internal |
| Herber et al.24 (2021) | Germany | KC to control | 126/93 | LDA & RF | Pentacam/10 | Internal |
| ∗Yousefi et al.25 (2018) | Japan | KC severity | 3156/- | DBC | CASIA-OCT/420 | NA |
Three classifications of KC diagnosis: Early KC from control, Clinical KC from control and KC severity. Multiple methods (MMs), Linear Discriminant Analysis (LDA), Neural Network (NN), Feedforward Neural Network (FNN), (Convolutional Neural Network (CNN), Random Forest (RF), Naïve Bayes (NB), Density-based clustering (DBC). (∗) Studies not included in the meta-analysis.