Multimodal convolutional neural network achieved highest predictive performance, with ROC-AUC of 0.90
By Elana Gotkine HealthDay Reporter
TUESDAY, Sept. 16, 2025 (HealthDay News) — An artificial intelligence (AI) algorithm can predict progression of keratoconus, according to a study presented at the annual Congress of the European Society of Cataract and Refractive Surgeons, held from Sept. 12 to 16 in Copenhagen, Denmark.
Shafi Balal, M.B.B.S., from Moorfields Eye Hospital in London, and colleagues used AI assistance to stratify the risk for keratoconus progression in a retrospective study using data collected between September 2019 and March 2023. Patients referred to the keratoconus clinic with multiple visits at least 90 days apart and at least six months of follow-up were included. Three AI models were developed: XBoost for tabular and demographic data; a unimodal convolutional neural network (CNN) for Placido and optical coherence tomography (OCT) images; and a multimodal CNN.
The dataset included 36,673 images from 6,684 eyes. The researchers found that 11.2 percent of patients had progression at any time point during follow-up within a mean follow-up of 18.9 months. The multimodal CNN achieved the highest predictive performance, with an area under the receiver operating characteristic curve (ROC-AUC) of 0.90. ROC-AUC values were 0.84 for combined modality approaches of OCT or Placido images with tabular data. When assessing single-modality approaches, the best performance was seen for Placido topography alone, followed by OCT imaging and tabular data (ROC-AUCs, 0.82, 0.79, and 0.78, respectively).
“Our results could mean that patients with high-risk keratoconus will be able to receive preventative treatment before their condition progresses,” Balal said in a statement. “The effective sorting of patients by the algorithm will allow specialists to be redirected to areas with the greatest need.”
Copyright © 2025 HealthDay. All rights reserved.
