Collaborative Feature and Persona Enhancement for Trustworthy Medical Foundation Models

Fanqi Kong, Ziming Zhao

Keywords: persona-aware foundation models; medical image segmentation; fairness and bias; trustworthy medical AI.

TL;DR: We introduce a persona-aware CEIGM-UNet that jointly models medical image segmentation and patient personas, achieving state-of-the-art accuracy with fewer parameters while improving fairness across demographic groups.

Abstract: Foundation models promise to democratize access to high–quality medical decision support by learning from vast quantities of data, but unmitigated biases in the data and model architecture can undermine their trustworthiness. Inspired by recent advances in persona–steered language modelling and efficient vision transformers, we propose a new architecture that jointly learns fine‑grained medical image representations and patient personas while accounting for fairness and cognitive plausibility. Our model builds upon a multi‑scale U‑shaped backbone with collaborative feature enhancement and Group Mamba layers. We introduce a persona module that conditions intermediate features on demographic embeddings and a psychologically motivated modulation function. Experiments on multi‑organ CT (Synapse) and cardiac MR (ACDC) benchmarks demonstrate competitive segmentation accuracy with substantially fewer parameters than conventional transformers. We further evaluate persona steerability and bias, showing that our approach produces more authentic persona behaviors than baseline methods while maintaining equitable performance across demographic groups. Finally, we discuss psychological foundations and ethical considerations of persona‑aware medical foundation models and outline directions for responsibly developing trustworthy AI in healthcare.

Submission Number: 139