ESCRS - FP06.14 - Ai-Driven Metrics For Predicting Subclinical Keratoconus Via Collagen Distribution Analysis With Ps-Oct

Ai-Driven Metrics For Predicting Subclinical Keratoconus Via Collagen Distribution Analysis With Ps-Oct

Published 2025 - 43rd Congress of the ESCRS

Reference: FP06.14 | Type: Free paper | DOI: 10.82333/d6qw-3h24

Authors: Ozana Moraru* 1 , Cristian Moraru 1 , Petru Pintea 1

1Anterior segment surgery,Oculus Eye Clinic,Bucharest,Romania

Purpose

This study aims to develop an AI-driven system for early keratoconus detection by analyzing collagen fibril orientation and phase retardation (PR) patterns using polarization-sensitive OCT (PS-OCT). By integrating AI-driven collagen distribution analysis with corneal topography, biomechanics (BM), and Bowman’s layer thickness (BLT), we propose an automated risk stratification tool to refine refractive surgery decision-making and prevent post-surgical ectasia.

Setting

A single-center prospective study conducted in a clinical research setting.

Methods

A total of 350 patients undergoing refractive surgery screening underwent corneal topography (Pentacam HR), BM analysis (Corvis-ST), and PS-OCT imaging. AI-based feature extraction analyzed PR maps, Bowman’s membrane integrity and thickness, and collagen fibril orientation. A machine learning model was trained using multimodal imaging data to generate a keratoconus likelihood score (0 = healthy, >0.5 = subclinical KC, >0.7 = clinical KC). AI dynamically weighted SpA1, CBI, PR distribution, and BLT to enhance diagnostic precision.

Results

  • AI-driven PR map analysis identified distinct collagen fibril patterns.
  • Normal corneas: Minimal retardation at the apex (21°), increasing peripherally (47°).
  • Subclinical KC: Fibril misalignment with significantly higher PR values (P<0.05).
  • AI-based risk stratification categorized eyes into:
    • Healthy (AI score <0.5): Uniform collagen distribution.
    • Subclinical KC (0.5–0.7): Early fibril disruption despite normal topography.
    • Clinical KC (>0.7): Structural abnormalities detected before clinical signs.
  • Strong correlations found between AI scores, SpA1, CBI, BLT, and epithelial remodeling.
  • AI improves early detection and enhances precision in keratoconus risk assessment.

Conclusions

AI-driven collagen analysis using PS-OCT enables early and precise keratoconus detection, refining refractive surgery planning and reducing ectasia risk. By integrating machine learning with multimodal imaging, this system provides an objective, automated risk stratification model, marking a paradigm shift in personalized ophthalmic diagnostics.