A polarization-sensitive optical coherence tomography (PS‑OCT) approach combined with artificial intelligence (AI) could help ophthalmologists refine the diagnosis of subclinical keratoconus and make safer refractive surgery decisions, according to a study published in Biophotonics Discovery.
The research, led by Rahul P. Patil and colleagues At the Narayana Nethralaya Foundation (India), evaluated whether ultrahigh-resolution PS-OCT could provide clinically useful information beyond conventional corneal tomography. Keratoconus diagnosis has advanced substantially with topography and tomography, but detecting subclinical keratoconus remains difficult because early disease may involve microstructural and biomechanical changes before obvious corneal deformation is visible.
PS-OCT offers a different view of the cornea by measuring phase retardation, a marker related to birefringence and stromal collagen organization. Because keratoconus is associated with disruption of the corneal collagen architecture, the authors investigated whether this technology could identify subtle disease patterns that might be missed or differently interpreted by standard imaging.
The retrospective study included 359 eyes from 197 subjects examined at Narayana Nethralaya Eye Hospital between January 2022 and September 2024. The cohort comprised 120 healthy eyes, 109 subclinical keratoconus eyes, and 130 keratoconus eyes. All participants underwent imaging with Pentacam, MS-39, and a custom-built ultrahigh-resolution PS-OCT device.
The investigators developed three random forest AI models, each trained on data from one imaging modality. The PS-OCT model used spatial maps of phase retardation, epithelial thickness, and Bowman’s layer thickness. The Pentacam and MS-39 models used device-specific keratometric, tomographic, epithelial, and aberration-derived parameters.
Across the full cohort, all three AI models performed well. Overall accuracy was 82 percent for PS-OCT and 86 percent for both Pentacam and MS-39. For clinically evident keratoconus, performance was particularly strong: PS-OCT and MS-39 correctly classified all 130 keratoconus eyes, while Pentacam correctly classified 129.
The most clinically interesting differences emerged in the subclinical group. PS-OCT classified 43 of 109 subclinical keratoconus eyes as healthy, compared with 30 for Pentacam and 33 for MS-39. In fellow eyes of highly asymmetric keratoconus cases, PS-OCT classified 21 of 33 as healthy, versus 12 for Pentacam and 16 for MS-39. At the same time, PS-OCT identified several subtle keratoconus cases that the tomography-based models classified as subclinical or uncertain.
The study authors suggest this bidirectional reclassification may reflect PS-OCT’s ability to capture stromal collagen organization and corneal sublayer changes rather than relying primarily on corneal shape.
The findings point to PS-OCT as a complementary tool rather than a replacement for established tomography. For clinicians, its potential value may lie in improving confidence around borderline cases, particularly before refractive surgery. However, the authors also stress that prospective longitudinal studies are now needed to determine whether PS-OCT-based reclassifications predict true disease progression.
