Assistance of Artificial Intelligence in Diagnosis of Vitreoretinal Lymphoma on Optical Coherence Tomography

Assistance of Artificial Intelligence in Diagnosis of Vitreoretinal Lymphoma on Optical Coherence Tomography

Vitreoretinal lymphoma (VRL) remains a diagnostic challenge due to its scarce prevalence, and delayed diagnosis usually results in blindness and even fatal outcomes. Herein, an artificial intelligence (AI) system is developed to identify VRL among 16 retinal diseases and conditions on optical cohere...

Full description

Saved in:
Bibliographic Details
Main Authors: Aidi Lin, Yuanyuan Peng, Tian Lin, Jun Dai, Jizhu Li, Tingkun Shi, Xixuan Ke, Xulong Liao, Danqi Fang, Man Chen, Huiyu Liang, Shirong Chen, Honghe Xia, Jingtao Wang, Zehua Jiang, Tao Li, Dan Liang, Shanshan Yu, Jing Luo, Ling Gao, Dawei Sun, Yih Chung Tham, Xinjian Chen, Haoyu Chen
Format: Article
Language:English
Published: Wiley 2025-04-01
Series:Advanced Intelligent Systems
Subjects:
artificial intelligence
few‐shot learning
meta‐learning
optical coherence tomography
vitreoretinal lymphoma
Online Access:https://doi.org/10.1002/aisy.202400500
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Vitreoretinal lymphoma (VRL) remains a diagnostic challenge due to its scarce prevalence, and delayed diagnosis usually results in blindness and even fatal outcomes. Herein, an artificial intelligence (AI) system is developed to identify VRL among 16 retinal diseases and conditions on optical coherence tomography (OCT) images with the cross‐subject meta‐transfer learning (CS‐MTL) algorithm. Extensive experiments of few‐shot VRL recognition tasks prove the robustness of our model on 1‐, 3‐, and 5‐shot scenarios, achieving an F1 score of 0.8697 to 0.9367. The superiority of the model is shown with a higher F1 score (0.9310) compared with other state‐of‐the‐art algorithms (0.5487–0.9018) and three doctors whose clinical experiences range between 3 to 10 years without the help of the CS‐MTL (0.7773–0.8949). AI assistance significantly improves the F1 scores of doctors by 6.16–14.46% (p < 0.001). Moreover, the F1 scores of AI‐assisted senior doctor and retinal specialist (0.9414 and 0.9500), but not the junior doctor (0.8897), exceed that of the CS‐MTL (0.9310). This study presents a promising approach for aiding in the diagnosis of VRL on retinal OCT images and may provide a novel insight into the collaboration of doctors with AI techniques, resulting in reducing the risk of diagnostic delays of rare diseases.
ISSN:2640-4567

Similar Items