Use of an Artificial Intelligence Model for Automated Measurement of the Anterior Segment by Ultrasound Biomicroscopy in ICL Calculation

PURPOSE: To evaluate the agreement between an artificial intelligence-based automated system and manual ultrasound biomicroscopy (UBM) measurements of anterior segment parameters relevant to Implantable Collamer Lens (ICL; STAAR Surgical) planning. METHODS: This cross-sectional, observational, and analytical study was conducted at a tertiary referral ophthalmology center in Mexico City, Mexico. A total of 400 ultrasound biomicroscopy images of the anterior segment corresponding to one image per patient were used to generate segmentation masks and train a U-Net-based automated measurement model. An independent test dataset consisting of 40 UBM images from 40 different patients, not included in the training dataset, was used for model evaluation. Dataset splitting was performed at the patient level to avoid data leakage. Automated measurements of anterior chamber depth (ACD), angle-to-angle distance (ATA), sulcus-to-sulcus distance (STS), ciliary body inner diameter (CBID), and sulcus-to-sulcus lens rise (SSLR) were compared with manual measurements performed by an experienced ocular ultrasonographer. RESULTS: Automated measurements showed good to excellent agreement with manual measurements for ACD, ATA, STS, and CBID (intraclass correlation coefficient [ICC]: 0.891 to 0.994). In contrast, SSLR demonstrated moderate agreement (ICC: 0.680 to 0.810). Bland-Altman analysis showed minimal bias and no evidence of proportional error across parameters. CONCLUSIONS: The artificial intelligence model demonstrated good agreement with manual measurements for most anterior segment parameters, with variable performance across parameters. These findings support the potential role of automated UBM-based measurements as an adjunctive tool for anatomical assessment in ICL planning. Further studies are warranted to determine its impact on ICL sizing decisions and postoperative outcomes.