Open-Source Multimodal Imaging and Clinical Data Set for Uveal Melanoma

To compile and release a publicly available, annotated, multimodal imaging data set for uveal melanoma (UM) and its clinical mimickers, addressing the current gap of open-source ophthalmic imaging resources needed for education and the development and validation of artificial intelligence (AI)-based diagnostic tools.

Retrospective, single-center data set compilation and descriptive study.

In total, 645 eyes from 645 patients diagnosed with UM, choroidal nevi, or congenital hypertrophy of the retinal pigment epithelium at the Illinois Eye and Ear Infirmary between January 2010 and February 2026.

Multimodal imaging data were retrospectively collected and paired for each eye, consisting of ultra-widefield fundus photographs and standardized B-scan ocular ultrasonography. Each case was annotated with structured data including patient demographics, tumor dimensions, and clinically relevant imaging features to support downstream computational analysis and studies.

Tumor dimensions (apical height and largest basal diameter) among UM cases, benchmarked against representative published cohorts, and diagnostic performance of prior AI models developed using this data set.

Among UM cases, mean apical tumor height was 5.63 ± 3.03 mm and mean largest basal diameter was 11.77 ± 2.90 mm, consistent with values reported in representative published cohorts. Computational studies using this data set have demonstrated utility for automated risk factor detection, tumor segmentation, and binary and multiclass lesion classification, with area under the receiver operating characteristic curve values of up to 0.98 achieved for UM vs. choroidal nevus discrimination.

To our knowledge, this is the first and largest publicly available, annotated, multimodal ophthalmic imaging data set for UM and its mimickers. Prior computational studies leveraging this data set have developed high-performing models across a variety of clinically relevant computer vision tasks, demonstrating the data set's utility for AI development and validation. This resource is intended to accelerate reproducible AI model development, support multicenter research collaboration, and enhance ocular oncology education.

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.