A specialized reasoning large language model for accelerating rare disease diagnosis: a randomized AI physician assistance trial
Summary
This paper presents RaDaR, a 32B open-source reasoning LLM trained on public and synthetic rare disease cases, which outperforms larger models like DeepSeek-R1 in diagnosis benchmarks and improves physician accuracy by 21.44 percentage points in a randomized trial.
View Cached Full Text
Cached at: 06/24/26, 07:47 AM
# A specialized reasoning large language model for accelerating rare disease diagnosis: a randomized AI physician assistance trial Source: [https://arxiv.org/abs/2606.24510](https://arxiv.org/abs/2606.24510) Authors:[Haichao Chen](https://arxiv.org/search/cs?searchtype=author&query=Chen,+H),[Songchi Zhou](https://arxiv.org/search/cs?searchtype=author&query=Zhou,+S),[Zhengyun Zhao](https://arxiv.org/search/cs?searchtype=author&query=Zhao,+Z),[Shikai Hu](https://arxiv.org/search/cs?searchtype=author&query=Hu,+S),[Xianghong Jin](https://arxiv.org/search/cs?searchtype=author&query=Jin,+X),[Hongwei Ji](https://arxiv.org/search/cs?searchtype=author&query=Ji,+H),[Li He](https://arxiv.org/search/cs?searchtype=author&query=He,+L),[Shuli Li](https://arxiv.org/search/cs?searchtype=author&query=Li,+S),[Yiming Qin](https://arxiv.org/search/cs?searchtype=author&query=Qin,+Y),[Xin Tan](https://arxiv.org/search/cs?searchtype=author&query=Tan,+X),[Runfeng Shi](https://arxiv.org/search/cs?searchtype=author&query=Shi,+R),[Yih Chung Tham](https://arxiv.org/search/cs?searchtype=author&query=Tham,+Y+C),[Jiaye Zhu](https://arxiv.org/search/cs?searchtype=author&query=Zhu,+J),[Ye Li](https://arxiv.org/search/cs?searchtype=author&query=Li,+Y),[Ye Jin](https://arxiv.org/search/cs?searchtype=author&query=Jin,+Y),[Longhao Cao](https://arxiv.org/search/cs?searchtype=author&query=Cao,+L),[Dawei Li](https://arxiv.org/search/cs?searchtype=author&query=Li,+D),[Honghan Wu](https://arxiv.org/search/cs?searchtype=author&query=Wu,+H),[Hongqiu Gu](https://arxiv.org/search/cs?searchtype=author&query=Gu,+H),[Guanqiao Li](https://arxiv.org/search/cs?searchtype=author&query=Li,+G),[Tudor Groza](https://arxiv.org/search/cs?searchtype=author&query=Groza,+T),[Chunying Li](https://arxiv.org/search/cs?searchtype=author&query=Li,+C),[Dian Zeng](https://arxiv.org/search/cs?searchtype=author&query=Zeng,+D),[Weihong Yu](https://arxiv.org/search/cs?searchtype=author&query=Yu,+W),[Gareth Baynam](https://arxiv.org/search/cs?searchtype=author&query=Baynam,+G),[Saumya Shekhar Jamuar](https://arxiv.org/search/cs?searchtype=author&query=Jamuar,+S+S),[Min Shen](https://arxiv.org/search/cs?searchtype=author&query=Shen,+M),[Shuyang Zhang](https://arxiv.org/search/cs?searchtype=author&query=Zhang,+S),[Bin Sheng](https://arxiv.org/search/cs?searchtype=author&query=Sheng,+B),[Sheng Yu](https://arxiv.org/search/cs?searchtype=author&query=Yu,+S),[Tien Yin Wong](https://arxiv.org/search/cs?searchtype=author&query=Wong,+T+Y) [View PDF](https://arxiv.org/pdf/2606.24510) > Abstract:Rare diseases affect millions of individuals worldwide, yet timely diagnosis remains a major public health challenge due to scarcity of specialized clinical expertise\. While large language models \(LLMs\) show promise to support rare disease diagnosis, current models are constrained by insufficient clinical deployability, limited clinically grounded evidence, and scarcity of training data\. Here we present RaDaR \(Rare Disease navigatoR\), an open\-source, compact reasoning LLM \(32B parameters\) for rare disease diagnosis\. RaDaR was trained with 49,170 publicly available free\-text cases and 104,666 synthetic cases with reasoning\-enhanced training\. RaDaR showed the strongest performance among evaluated open\-source models, including the 671B DeepSeek\-R1, across public benchmarks and four external validation centers\. In a retrospective cohort, RaDaR prioritized the final diagnosis before documented clinical suspicion in 61\.06 percent of cases, corresponding to a potential lead time of 1\.87 months and 50\.18 percent of the within\-center interval\. In a randomized physician\-assistance trial, RaDaR assistance improved physicians' rare\-disease diagnostic accuracy by 21\.44 percentage points compared with internet search alone\. Synthetic\-data ablations suggested that phenotype\-anchored narratives provide useful training signal for long\-tail rare diseases, with a monotonic scaling trend within the tested data range\. Together, RaDaR and its development and validation framework provide a deployable rare\-disease reasoning model and a reproducible development framework for diagnostic AI under data scarcity\. ## Submission history From: Haichao Chen \[[view email](https://arxiv.org/show-email/914669c4/2606.24510)\] **\[v1\]**Tue, 23 Jun 2026 12:42:23 UTC \(1,748 KB\)
Similar Articles
RareDxR1: Autonomous Medical Reasoning for Rare Disease Diagnosis Beyond Human Annotation
Introduces RareDxR1, an end-to-end reasoning-centric large language model for open-domain rare disease diagnosis from unstructured clinical notes, using a progressive training framework and reflection-enhanced reasoning sampling, achieving state-of-the-art accuracy.
Using AI to help physicians diagnose rare genetic diseases affecting children
Researchers from Boston Children's Hospital, Harvard, and OpenAI used the OpenAI o3 Deep Research reasoning model to reanalyze 376 unsolved rare disease cases, leading to diagnoses in 18 additional cases (4.8% yield) after expert review and clinical confirmation. The study, published in NEJM AI, demonstrates how AI-assisted workflows can help experts revisit difficult cases as scientific knowledge evolves.
Reinforcement Learning for Evidence-Seeking Diagnostic Reasoning with Large Language Models
This paper proposes a reinforcement learning framework for evidence-seeking diagnostic reasoning using LLMs. The RL-trained 7B model outperforms larger models in multilingual clinical consultation tasks, showing that specialized RL can distill high-level clinical reasoning.
DeepLens Diagnosis Agent: Agentic Workflow Design Lets a Small Reasoning Model Compete with Frontier LLMs
The DeepLens Diagnosis Agent uses a five-stage agentic workflow with a small medical reasoning model (7B) to achieve 60.14% diagnostic accuracy on a 915-case benchmark, outperforming frontier LLMs like Claude Sonnet 4.5 and Gemini 3.1 Pro at lower cost. The workflow design alone yields a 36-point gain over the base model, demonstrating that structured process constraints are key for diagnostic reasoning.
@OpenAI: Rare disease diagnosis is challenging, as sequencing can surface millions of variants, and medical knowledge changes co…
OpenAI highlights how o3 Deep Research can aid rare disease diagnosis by integrating clinical features, inheritance patterns, variant evidence, and scientific literature into actionable hypotheses for specialists.