Rev Bras Oftalmol.2026;85:e0105
The enhanced pachymetry-adjusted/inferior-superior index for keratoconus detection with artificial intelligence: a tribute to Dr. Yaron Rabinowitz
DOI: 10.37039/1982.8551.20260105
ABSTRACT
Objective:
To pay a scientific tribute to Dr. Yaron Rabinowitz by enhancing his PA/I-S index through machine learning-based predictive models using the Applied Artificial Intelligence (A²I) methodology.
Methods:
A retrospective study with 2,705 eyes (1,680 normal and 1,025 VAE) to develop a predictive model (AI Tribute) following the KDD knowledge discovery framework, whose performance was assessed in two clinical scenarios – Normal versus Clinical Ectasia (KC+VAE-E) (N=3,335) and Normal versus VAE-NT (N=2,231) – by AUC and the DeLong test.
Results:
In the Normal versus Clinical Ectasia scenario, AI Tribute achieved an AUC of 0.985, significantly outperforming PA/I-S index (0.978), I-S value (0.968), KISA% (0.953), and pachy min (0.938). In the Normal versus VAE-NT scenario, AI Tribute achieved an AUC of 0.791, remaining superior to all reference indices.
Conclusion:
Machine learning modeling of Dr. Rabinowitz’s foundational variables significantly improved keratoconus detection across both clinical scenarios, demonstrating that the A²I approach amplifies established clinical knowledge even in the challenging task of detecting topographically normal ectatic eyes.
