CausalFund: Causality-Inspired Domain Generalization in Retinal Fundus Imaging for Low-Resource Screening

Abstract Early screening for glaucoma and diabetic retinopathy (DR) is critical to prevent irreversible vision loss, yet remains inaccessible to many underserved populations. However, AI models trained on hospital-grade fundus images often generalize poorly to low-cost images acquired with portable devices such as smartphones. We proposed CausalFund, a causally inspired learning framework for training AI models that enable reliable low resource screening from easily acquired non-clinical images. CausalFund disentangles disease-relevant retinal features from spurious image factors to achieve domain generalizable screening across clinical and nonclinical settings. We integrated CausalFund with seven deep learning backbones for glaucoma and DR screening from portable-device fundus images, including lightweight architectures suitable for on-device deployment. Across diverse experimental settings and image quality conditions, CausalFund improved smartphone-domain AUC for most backbones while maintaining comparable hospital-domain performance, and achieved a more balanced sensitivity-specificity trade-off than baselines. As a model-agnostic framework, CausalFund could be extended to other diseases and low-resource settings characterized by degraded or non-standard imaging.