New Publication(ETF-UML for incomplete clinical information fusion) in Information Fusion

We are pleased to announce that our team has published a new paper in Information Fusion (Impact Factor: 17.4), one of the leading journals in the field of information fusion and multimodal learning.

This study addresses clinical prediction as an incomplete information fusion problem, where the set of observable modalities—coded patient context, lab or vital trajectories, and free-text documentation—varies across care workflows and observation timepoints. The proposed framework, ETF-UML, trains modality-specific encoders independently while aligning them to a shared Equiangular Tight Frame (ETF) decision geometry, and then composes only the observed modalities through metadata-conditioned fusion. This design enables robust unimodal-to-multimodal composition when modality acquisition is asynchronous, incomplete, and time-varying. Evaluated on MIMIC-IV 30/90-day mortality and eICU 10-day in-hospital mortality prediction, the model shows consistent gains in AUROC and average performance over end-to-end fusion and missing-modality baselines, with additional experiments attributing the gains to the sharing and the fixing of the decision geometry, including on a task with more than two classes and under reduced modality availability.

"Geometry-guided unimodal-to-multimodal composition for incomplete and asynchronous clinical information fusion" JaeSeong Hong, Mingyu Kim, & Yu Rang Park (2026). Information Fusion.