Enhancing corneal ectasia susceptibility detection: analysis of a new algorithm (BAD-D v4)

Abstract Accurate detection of post-refractive ectasia susceptibility is essential during preoperative evaluation for laser vision correction (LVC) due to the risk of progressive corneal ectasia and vision decline post-surgery. Despite improved screening and a reduced incidence from 0.66 to 0.033%, iatrogenic ectasia remains a concern due to the severe vision loss it can cause, highlighting the need for more accurate detection tools. A new optimized version of the Belin/Ambrósio Enhanced Ectasia Display version 4 (BAD-D v4) was developed and validated across 26 international centers to enhance the detection of keratoconus and very asymmetric ectasia and to assess the risk of post-refractive ectasia. Analyzing a dataset of 3,886 eyes from 3,351 patients, including normal, keratoconus (KC), and cases with very asymmetric ectasia (VAE) categories, having one eye with normal topography (VAE-NT and the fellow eye with clinical ectasia (VAE-E). The study utilized an optimized logistic regression algorithm improving diagnostic accuracy. The BAD-D v4 showed superior efficacy in differentiating normal eyes from ectatic conditions, with Area Under the Receiver Operating Characteristic Curve (AUROC) scores of 0.997 and 0.998 in training and testing samples for normal versus clinical ectasia. Additionally, in Normal vs. Disease (KC + VAE), the AUROC was 0.974 and 0.966, and in the challenging Normal vs. VAE-NT diverse group, it scored 0.905 and 0.858. These results outperformed the current version (BAD-D v3) and were comparable to the Pentacam Random Forest Index in all tested scenarios, highlighting the potential of BAD-D v4 in early ectasia detection, without altering the index scale or the end-user experience.