ScholarDeck

AI / informaticsOphthalmology science · 2026

Attention-Based Multimodal Deep Learning for Uveal Melanoma Classification Using Ultra-Widefield Fundus Images and Ocular Ultrasound.

Albert K Dadzie, Sabrina P Iddir, Mansour Abtahi … Xincheng Yao

OphthalmologyAI / informatics

Combining fundus photos and ultrasound with attention outperformed either alone for melanoma classification.

Attention-Based Multimodal Deep Learning for Uveal Melanoma Classification Using Ultra-Widefield Fundus Images and Ocular Ultrasound.

Albert K Dadzie … Xincheng Yao

Ophthalmology science · 2026

Purpose

To develop and evaluate a deep learning model that integrates ultra-widefield fundus photography and B-scan ultrasonography for automated classification of uveal melanoma (uveal melanoma) and choroidal nevi.

Methods

A retrospective cross-sectional study.

n = 174 patients

0.9606

Results

0.9606

combining both scans, not just one, most reliably told melanoma from harmless nevi

n = 174 patients

More results

Uveal melanomas had a mean thickness of 6.0 mm and a basal diameter of 12.6 mm, whereas nevi measured 1.8 mm and 6.5 mm, respectively.

Among single-modality models, the model trained on transverse ultrasound images achieved the highest performance (accuracy: 92%; F1 score: 0.9227; AUC: 0.9538).

More results

Averaging predictions from the single-modality models provided only modest gains because their outputs sometimes conflicted.

“

Conclusion · 1 of 2

Multimodal deep learning that combines fundus photography and ultrasound imaging improves the classification of uveal melanoma and choroidal nevi.

Conclusion · 2 of 2

This approach demonstrates feasibility for leveraging the strengths of each modality for automated classification of uveal melanoma and choroidal nevi.

Post

Attention-Based Multimodal Deep Learning for Uveal Melanoma Classification Using Ultra-Widefield Fundus Images and Ocular Ultrasound.

Combining fundus photos and ultrasound with attention outperformed either alone for melanoma classification.

Related papers