A Multi-Center Benchmark for Abdominal Disease Diagnosis and Report Generation from Non-Contrast CT
中文标题: 基于平扫CT的腹部疾病多中心基准：诊断与报告生成
英文摘要
This paper introduces a multi-center benchmark for multi-organ abdominal disease diagnosis and automated radiology report generation using only non-contrast CT (NCCT), aiming to eliminate contrast agent risks. A large-scale paired NCCT-CECT dataset with corresponding reports was curated from two centers, split into internal and external cohorts. Five contemporary deep learning architectures, including chest-specific, abdomen-specific, and general-purpose multimodal models, were evaluated under a unified protocol. NCCT-based models achieved an average multi-organ AUC of 69.1% on the internal set and 63.1% on the external set, demonstrating retained diagnostic signals. The authors release the dataset, code, and benchmark publicly to advance contrast-free abdominal imaging research.
中文摘要
该论文提出一个多中心基准，仅利用平扫CT进行多器官腹部疾病诊断和自动化放射报告生成，旨在避免造影剂相关风险。研究收集了来自两个中心的配对平扫-增强CT数据集及相应报告，划分为内部和外部验证队列。在统一评估协议下，对五种当代深度学习架构（涵盖胸部专用、腹部专用和通用多模态模型）进行了基准测试。基于平扫CT的模型在内部队列上平均多器官AUC为69.1%，外部队列为63.1%，表明平扫CT保留了诊断信号。作者公开了数据集、代码和基准，以促进安全、资源高效的无造影腹部影像研究。
关键要点
First multi-center benchmark for abdominal disease diagnosis and report generation from non-contrast CT, synthesizing contrast-enhanced findings.
首个基于平扫CT进行腹部疾病诊断和报告生成的多中心基准，旨在合成增强CT所见。
Curated a large paired NCCT-CECT dataset with radiology reports across two centers, including internal and external validation sets.
从两个中心收集了大型配对平扫-增强CT数据集及放射报告，含内部和外部验证集。
Five deep learning architectures (chest-specific, abdomen-specific, multimodal) were benchmarked, showing NCCT can retain diagnostic signals.
对五种深度学习架构（胸部专用、腹部专用、多模态）进行基准测试，证明平扫CT可保留诊断信号。
NCCT-based models achieved average multi-organ AUC of 69.1% (internal) and 63.1% (external), confirming feasibility for contrast-free workflows.
基于平扫CT的模型内部平均多器官AUC 69.1%，外部63.1%，验证了无造影剂工作流的可行性。
Dataset, code, and evaluation protocol are publicly released to spur safer, globally accessible abdominal imaging research.
公开数据集、代码与评估协议，推动更安全、全球可及的无造影剂腹部影像研究。