# AI-Based Estimate of the Regional Effect of Orthokeratology Lenses on Tear Film Quality

## TL;DR

This study uses AI to detect subtle changes in tear film quality caused by orthokeratology lenses, revealing issues missed by traditional methods.

## Contribution

The novel use of AI and regional spatial mapping reveals previously undetected tear film changes in Ortho-K lens wearers.

## Key findings

- Regional tear film deterioration occurs in nasal and temporal corneal zones over time.
- Global mean metrics fail to capture these subtle regional changes.
- An AI model accurately predicts spatial tear film quality with high precision.

## Abstract

Purpose: To investigate regional changes in tear film quality associated with orthokeratology (Ortho-K) lens wear using high-resolution spatial mapping and to evaluate the potential of artificial intelligence (AI) models in anticipating these changes. Methods: This study analysed tear film quality in 92 Ortho-K wearers divided into three groups based on lens wear duration (10–29 days, 30–90 days, and ≥91 days). Placido-based topographer measurement was used to generate regional tear film maps before and after treatment. A custom MATLAB pipeline enabled regional comparisons and statistical mapping. A feedforward neural network was trained to forecast local tear film quality using spatial data. Results: Single-value global mean metrics showed minimal changes in tear film quality across groups. However, regional mean mapping revealed significant mid-peripheral and peripheral deterioration over time, particularly in nasal and temporal corneal zones. These changes were often overlooked by global averaging and remained invisible through tear film breakup time (TBUT) measurements. The AI model predicted spatial tear quality with high accuracy (R ≥ 0.9 in testing), capturing nuanced regional variations. Conclusions: The regional analysis uncovers subtle, clinically relevant tear film disruptions caused by Ortho-K lens wear, particularly in peripheral areas. These insights challenge the adequacy of traditional single-value global mean assessments. The AI model demonstrates the potential for non-invasive, predictive evaluation of tear stability, supporting more personalised and effective Ortho-K care.