Who Should Be Consulted? Targeted Expert Selection for Rare Disease Diagnosis

Yinghao Fu, Chao Yang, Xinye Chen, Yuting Yan, Shuang Li

Keywords: Human-AI Complementarity

Abstract: Effective collaboration between human experts and AI systems holds great promise in enhancing complex decision-making, particularly in challenging domains like *rare disease diagnosis*. In traditional Multi-Disciplinary Team (MDT) settings, human experts from different specialties are pre-assigned to review and discuss patient cases collaboratively. However, such fixed team structures may suffer from cognitive anchoring, incomplete knowledge, or misaligned expertise, especially when facing atypical or rare clinical presentations. In this paper, we propose **S**equential **E**xpert **E**ngagement for **R**are diseases (SEER) that *dynamically selects targeted human experts to participate in collaborative decision-making*. Our approach leverages a rule-based AI system with broad, structured medical knowledge to identify critical diagnostic paths and propose complementary expert inputs. The AI system serves two key roles: (1) recommending plausible diagnostic hypotheses and logical rules based on structured knowledge; and (2) identifying which experts, if consulted, are most likely to resolve diagnostic uncertainties. This targeted expert selection process helps avoid cognitive biases like anchoring and expands the decision space by inviting diverse, high-value perspectives. Moreover, the system is self-evolving, continuously updates its rules and its understanding of each expert's expertise based on newly collected data and feedback. Experiments on both synthetic and real-world rare disease datasets demonstrate that our framework improves diagnostic accuracy, reduces expert workload, and enhances the overall robustness of human-AI collaboration.

Submission Number: 29