ASTAR - Automated induction of STAndardized radiology Reporting templates from large-scale clinical free-text corpora

Sep 15, 2026

Upload by Quinn

Structured reporting converts free-text radiology narratives into queryable data keys, enabling cohort assembly, longitudinal tracking, and training-label generation for medical AI. Current approaches rely on a two-stage pipeline where information extraction has improved with large language models, but template construction remains a manual bottleneck based on labor-intensive, static expert consensus. ASTAR is an LLM-based framework that induces standardized reporting templates from large-scale clinical free-text corpora. Experiments on 4,215 fetal brain MRI reports show superior coverage, fidelity, and usability, reducing template development from weeks to hours.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Structured reporting and its value

Two-stage pipeline and existing limitations

Method

Overview of ASTAR

Automated template induction

arXiv : 2608 .20369v 1 [cs .CL] 19 Jun 2026 ASTAR: Automated induction of STAndardized radiology Reporting templates from large-scale clinical free-text corpora Xinfeng Zhang 1†, Mingxuan Liu 1†, Yifei Chen 1 , Juncheng Zhu2 , Kasidit Anmahapong 1 , Yiming Huang3 , Yuan Zhang4 , Hongjia Yang 1 , Yi Liao2 , Gang Ning2 , Haibo Qu2 , and Qiyuan Tian 1 ∗ 1 Tsinghua University, Beijing, China 2 Sichuan University, Chengdu, China 3 University of California San Diego, San Diego, USA 4 Southeast University, Nanjing, China [qiyuantian@tsinghua.edu.cn](qiyuantian@tsinghua.edu.cn) Abstract. Structured reporting converts free-text radiology narratives into queryable data keys, facilitating cohort assembly, longitudinal tracking, and training label generation for medical AI. The prevailing paradigm follows a two-stage pipeline: (1) constructing a reporting template, (2) extracting information to populate it. While the extraction stage has benefited from advances in large language models (LLMs), template construction remains a manual bottleneck relying on labor-intensive expert consensus that is static, difficult to scale, and may fail to capture real-world reporting diversity. We address this limitation with ASTAR, an LLM-based framework for Automated induction of STAndardized radiology Reporting templates from large-scale clinical free-text corpora. Extensive experiments on 4,215 fetal brain MRI reports from multiple centers demonstrate that the ASTAR-induced template surpasses two expert-curated templates across template coverage, information fidelity, diagnostic fidelity, and expert-rated usability, reducing template development from weeks of committee deliberation to hours of automated processing. Code: [https://github.com/birthlab/ASTAR](https://github.com/birthlab/ASTAR) Keywords: Radiology report · LLM · Structure information extraction. 1 Introduction Radiology departments worldwide generate clinical reports at an ever-growing scale. In the United States alone, over 16 . 1 million CT examinations were reported during a nine-month period in 2020 [11], and imaging utilization across all modalities is projected to rise 17–27% by 2055 [7] . Each examination produces a detailed free-text narrative synthesizing categorical descriptors, quantitative measurements, and diagnostic impressions, thereby forming one of the richest † Equal contribution. ∗ Corresponding author. 2 X. Zhang et al. Fig. 1. (A) Traditional template construction pipeline. (B) Our automated template induction pipeline (ASTAR) . data repositories in modern healthcare [17,5] . However, the utility of these reports is constrained because findings remain archived as unstructured text with wide stylistic, terminological, and structural variations across radiologists, institutions, and languages [15,16] . Such heterogeneity hinders systematic case retrieval, cohort assembly, longitudinal tracking, and training-label generation for medical AI [6,17] . Radiology societies therefore increasingly advocate for structured reporting [20,4], converting narratives into queryable keys that support rapid cohort identification and high-quality ground-truth generation [1,2] . Nevertheless, achieving these benefits at scale requires automated tools, since manual structuring of the massive and continuously growing volume of narrative reports is highly impractical. Therefore, recent studies increasingly employ natural language processing (NLP) and large language models (LLMs) for automated information extraction [12,23] . Specifically, the prevailing paradigm follows a two-stage pipeline: (1) a reporting template (i.e., a target template defining theset of keys and their permissible values) is constructed; and (2) relevant information is extracted from free-text reports to populate this template. For example, BURExtract-Llama [6] extracts clinical concepts from breast ultrasound reports with an average F1 score of 84 .6% . However, while the extraction st

FAQs

What problem does ASTAR address in structured radiology reporting?

It targets the manual bottleneck in constructing standardized reporting templates, which limits scalability and fails to reflect real-world reporting diversity.

How does ASTAR differ from the typical two-stage pipeline?

Instead of relying on expert-designed templates, ASTAR induces a hierarchical reporting template automatically using an LLM-based framework.

What evidence supports ASTAR’s effectiveness?

Experiments on 4,215 fetal brain MRI reports from multiple centers show ASTAR’s induced template outperforms expert-curated templates in coverage, information fidelity, diagnostic fidelity, and usability.

Sep 15, 2026

Upload by Quinn