Artificial neural network to guide intracorneal ring segments implantation for keratoconus treatment: a pilot study

Abstract

Background: To analyze the clinical results of an artificial neural network (ANN) that has been processed in order to improve the predictability of intracorneal ring segments (ICRS) implantation in keratoconus. Methods: This retrospective, comparative, nonrandomized, pilot, clinical study included a cohort of 20 keratoconic eyes implanted with intracorneal ring segments KeraRing (Mediphacos, Belo Horizonte, Brazil) using the ANN (ANN group) and 20 keratoconic eyes implanted with KeraRing using the manufacturer's nomograms (nomogram group). Uncorrected distance visual acuity (UDVA), corrected distance visual acuity (CDVA) (visual acuity is expressed in decimal value and in LogMAR value in brackets), manifest refraction, corneal topography, tomography, aberrometry, pachymetry and volume analysis (Sirius System. CSO, Firenze, Italy) were performed during the preoperative visit; and the two groups, ANN group and nomogram group, did not differ significantly preoperatively in all of the parameters evaluated. These preoperative values were compared with the results obtained at the third-month visit. Mann-Whitney test and Wilcoxon test were used for the statistical analyses. Results: The spherical equivalent and the keratometric values decreased significantly in both groups. The CDVA improved from 0.60 ± 0.23 (0.22 LogMAR) pre-operatively to 0.73 ± 0.21 (0.14 LogMAR) post-operatively in the ANN group (p < 0.005), and from 0.54 ± 0.19 (0.27 LogMAR) pre-operatively to 0.62 ± 0.19 (0.21 LogMAR) post-operatively in the nomogram group (p < 0.01), with statistically significant difference between the two groups (p < 0.05), being better in the ANN group. Coma-like aberrations decreased significantly in the ANN group, while in the nomogram group they did not change significantly, but no statistically significant difference was found between the two groups. Conclusions: ANN to guide ICRS provides an increase in the visual acuity, reduction in the spherical equivalent and improvement in the optical quality of keratoconus patients. ANN gives better results when compared with the manufacturer's nomograms in terms of better corrected vision and reduction of the coma-like aberrations. The constant inclusion of new cases will make the predictability of ANN increasingly better as the software finetunes its learning.

Keratoconus: An updated review

“…Epi-on CXL, the use of Meran riboflavin, and applying the accelerated irradiation protocol appeared to be associated with reduced efficacy regarding controlling keratoconus progression, with one-third of cases treated using epi-on CXL required re-treatment. Corneal cross-linking has also been used successfully in combination with corneal ring segments and other surgical techniques [366][367][368][369][370][371]. 9.3.1.2.…”

Detecting Keratoconus From Corneal Imaging Data Using Machine Learning

“…Moreover, this study used a relatively small sample. Machine learning has been also applied in keratoconus management with regard to guiding intra-corneal ring segment implantation [12]. This is promising and shows that AI models can be applied to different aspects of keratoconus management to enhance care delivery.…”

Comparative analysis of two different types of intracorneal implants in keratoconus: A corneal tomographic study

“…[2][3][4] ICRS improve visual acuity and ultimately life quality through corneal remodeling and providing a more regularized corneal surface. 5,6 Ring segment implanted in the midperipheral corneal area flatten the central zone due to an arc-shortening effect, and consequently reduce the myopic refraction. 7,8 Two different designs of intrastromal implants have been developed, including ring segments and complete or continuous rings.…”