| Unique ID issued by UMIN | UMIN000063691 |
|---|---|
| Receipt number | R000072814 |
| Scientific Title | Automated Structuring of Diagnostic Imaging Reports: Risk Management Using Large Language Models |
| Date of disclosure of the study information | 2026/11/19 |
| Last modified on | 2026/11/19 13:17:05 |
Automated Structuring of Diagnostic Imaging Reports: Risk Management Using Large Language Models
Automated Structuring of Diagnostic Imaging Reports: Risk Management Using Large Language Models
Automated Structuring of Diagnostic Imaging Reports: Risk Management Using Large Language Models
Automated Structuring of Diagnostic Imaging Reports: Risk Management Using Large Language Models
| Japan |
All conditions undergoing diagnostic imaging
| Radiology |
Others
NO
The objective of this study is to optimize closed large language models (LLMs) through prompt engineering and related techniques to evaluate the feasibility of a medical safety support system based on diagnostic imaging reports. Specifically, we aim to verify: 1) the inference speed and accuracy of automated structuring of free-text diagnostic imaging reports; 2) the accuracy of automated detection models for high-risk cases requiring follow-up, such as emergency conditions and incidental findings; and 3) the usefulness of a risk management system that cross-references radiology reports with electronic health records (HIS) to track clinical follow-up actions and detect unaddressed cases.
Others
Evaluation of structuring accuracy of diagnostic imaging reports and feasibility of a risk case detection system
Not applicable
Technical accuracy of automated structuring and high-risk case extraction from diagnostic imaging reports (accuracy, precision, recall, F1 score, sensitivity, and specificity compared with ground truth labeled by board-certified radiologists) (at the time of data analysis)
Observational
| Not applicable |
| Not applicable |
Male and Female
Finalized diagnostic imaging reports interpreted by board-certified radiologists for patients who underwent CT, MRI, X-ray (X-P), or radioisotope (RI) examinations at St. Marianna University Hospital between January 1, 2020, and December 31, 2024.
Diagnostic imaging reports of patients who opted out (declined participation) of the use of their data for this study.
10000
| 1st name | Yasuyuki |
| Middle name | |
| Last name | Kobayashi |
St. Marianna University School of Medicine
Department of Radiology
216-8511
2-16-1 Sugao, Miyamae-ku, Kawasaki, Kanagawa 216-8511, Japan
044-977-8111
yasukoba@marianna-u.ac.jp
| 1st name | Yasuyuki |
| Middle name | |
| Last name | Kobayashi |
St. Marianna University School of Medicine
Department of Advanced Biomedical Imaging Informatics
216-8511
2-16-1 Sugao, Miyamae-ku, Kawasaki, Kanagawa 216-8511, Japan
044-977-8111
yasukoba@marianna-u.ac.jp
St. Marianna University School of Medicine
St. Marianna University School of Medicine
Other
St. Marianna University School of Medicine
2-16-1 Sugao, Miyamae-ku, Kawasaki, Kanagawa 216-8511, Japan
044-977-8111
yasukoba@marianna-u.ac.jp
NO
| 2026 | Year | 11 | Month | 19 | Day |
Unpublished
Preinitiation
| 2026 | Year | 04 | Month | 10 | Day |
| 2026 | Year | 12 | Month | 01 | Day |
| 2028 | Year | 03 | Month | 31 | Day |
This is a single-center, retrospective observational study. We will collect and analyze existing data from approximately 10,000 finalized diagnostic imaging reports interpreted by board-certified radiologists (covering CT, MRI, X-ray [X-P], and radioisotope [RI] examinations) and corresponding electronic health records (clinical notes, order histories, and laboratory test results) for patients examined at St. Marianna University Hospital between January 1, 2020, and December 31, 2024. A unique study ID will be assigned to each patient for longitudinal analysis, and data stratified by modality and anatomical region will be divided into training and independent validation datasets. Using lightweight large language models (LLMs) operating in an on-premises closed environment, we will perform automated report structuring and high-risk case extraction, evaluating technical accuracy and clinical utility against ground truth labeled by board-certified radiologists and actual clinical follow-up actions recorded in electronic health records.
| 2026 | Year | 11 | Month | 19 | Day |
| 2026 | Year | 11 | Month | 19 | Day |
Value
https://center6.umin.ac.jp/cgi-open-bin/ctr_e/ctr_view.cgi?recptno=R000072814