Concept-guided noisy negative suppression during vision-language contrastive learning with hierarchical ontology extracted by LLMs from radiology report.

An evidential reasoning framework for interpretable real-world disease screening, leveraging regional evidence from historical cases retrieved via dual knowledge banks.

A comprehensive benchmark for retinal anomaly detection, covering data, algorithms, generalization evaluation, and a new NFM-DRA baseline.

Efficient medical vision-language alignment by adapting masked vision models with temporal-multiview radiograph inputs.

Masked record modeling for knowledge-enhanced semantic representations.

Using registration network to enhance image-to-image translation.