Corneal Topography Data for Explainable AI in Keratoconus Detection

A prospective pilot cohort of 20 eyes was used to develop an explainable neuro-symbolic AI framework for corneal diagnostics. The dataset, authored by Mini Han Wang and last updated in April 2026, integrates quantitative biometric parameters and corneal curvature maps from IOLMaster 700 reports. It is a 13.7 KB document describing the methodology and results of a system for automated interpretation and early disease detection.

Use Cases

• Training hybrid CNN-Vision Transformer models for spatial corneal morphology analysis based on corneal curvature maps.
• Building or validating a symbolic corneal knowledge graph for probabilistic reasoning based on relationships between corneal parameters and disease states.
• Developing large language model (LLM) modules for generating structured, bilingual clinical reports to explain AI reasoning.
• Benchmarking diagnostic performance for early keratoconus detection based on AUC, sensitivity, and specificity metrics.

Strengths

• The proposed framework achieved a diagnostic AUC of approximately 0.95 in the pilot cohort.
• The end-to-end pipeline processed a case in approximately 95 ± 12 seconds, supporting near-real-time use.
• Expert evaluation indicated high interpretability and clinical usefulness of the generated reports.
• Released under a permissive CC-BY-4.0 license.

Limitations

• Row count is unknown, which may limit suitability assessment.
• The current results are based on a pilot cohort of only 20 eyes, limiting statistical power and generalizability.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

figshare, authored by Mini Han Wang.

Collection Method

Data extracted from IOLMaster 700 reports for a prospective pilot cohort.

Time Range

Temporal coverage of the underlying data is not specified.

Freshness

Last updated 2026-04-13 05:46:43; freshness should be verified.

Geography

Spatial coverage is not specified.