Photo Credit: yogenyogeny

The following is a summary of “Detection of Ocular Surface Squamous Neoplasia Using Artificial Intelligence with Anterior Segment Optical Coherence Tomography,” published in the February 2025 issue of American Journal of Ophthalmology by Greenfield et al. 

Researchers conducted a retrospective study to develop and validate a deep learning (DL) model for distinguishing ocular surface squamous neoplasia (OSSN) from pterygium and pinguecula using high-resolution anterior segment optical coherence tomography (AS-OCT). 

They extracted imaging data from Optovue AS-OCT (Fremont, CA) and collected clinical or biopsy-confirmed diagnoses from electronic medical records. A DL model was developed using 2 approaches: (1) a masked autoencoder trained on 1,05,859 unlabeled AS-OCT images from 5,746 eyes and (2) a Vision Transformer supervised model fine-tuned with labeled data for binary classification (OSSN vs non-OSSN). Expert graders classified 2,022 AS-OCT images from 523 eyes (427 patients) into “OSSN or suspicious for OSSN” and “pterygium or pinguecula.” Diagnostic performance was tested on 566 scans (62 eyes, 48 patients) with biopsy-proven OSSN and compared with masked expert clinicians at the scan level without access to clinical images or data. 

The results showed that the DL model achieved an accuracy of 90.3% (95% CI: 87.5-92.6%), with a sensitivity of 86.4% (95% CI: 81.4-90.4%) and a specificity of 93.2% (95% CI: 89.9-95.7%) compared to biopsy-proven diagnoses. Expert graders had a lower sensitivity of 69.8% (95% CI: 63.6-75.5%) but a slightly higher specificity of 98.5% (95% CI: 96.4-99.5%). The area under the receiver operating characteristic curve (AUC) for the DL model was 0.945 (95% CI: 0.918-0.972), which was significantly higher than the AUC for expert graders (AUC = 0.688, P < 0.001). 

Investigators concluded that a DL model applied to AS-OCT scans demonstrated high accuracy in differentiating OSSN from pterygium and pinguecula, performing comparably to expert clinicians and showing promise for enhancing clinical decision-making. 

Source: ajo.com/article/S0002-9394(25)00078-9/abstract