Artificial Intelligence-Assisted Segmentation and Characterization of Choroidal Melanoma Using Ocular Ultrasound.
Authors
Affiliations (5)
Affiliations (5)
- Department of Quantitative Methods, Universidad Loyola Andalucía, Seville, Spain. [email protected].
- Department of Ophthalmology, Eye Cancer Institute, San Pedro Garza García, Nuevo León, Mexico.
- Department of Computer and Industrial Engineering, University of Monterrey, San Pedro Garza García, Nuevo León, Mexico.
- Department of Engineering, University of Monterrey, San Pedro Garza García, Nuevo León, Mexico.
- Department of Psychology, University of Monterrey, San Pedro Garza García, Nuevo León, Mexico.
Abstract
Choroidal melanoma is the most common primary intraocular malignancy in adults, and its evaluation relies heavily on ocular ultrasound, where interpretation is often based on manual analysis and subject to operator-dependent variability. This study presents an automated system for the analysis of B-scan ultrasound images of clinically confirmed choroidal melanoma cases, focusing on tumor segmentation, thickness measurement, and echogenicity classification. The proposed solution integrates image processing techniques and artificial intelligence (AI) models to enable consistent and reproducible extraction of clinically relevant parameters. The algorithm was developed following a review of state-of-the-art methods and validated using real-world ultrasound data, comparing its results with measurements performed by experienced ophthalmologists. The system was implemented within a web-based platform that allows users to upload ultrasound reports in PDF format. The ultrasound images are automatically extracted from the report, after which the user selects the image to be analyzed. The proposed system then automatically obtains quantitative measurements to support clinical evaluation. To ensure scalability and accessibility, the platform was deployed in a cloud-based environment. Given that B-scan ultrasound remains the most widely used imaging modality in choroidal melanoma [1], this work introduces a practical and scalable approach to standardize ultrasound-based assessment, reduce operator-dependent variability, and support clinical decision-making in ocular oncology.