Page 91 - AIH-1-4

Artificial Intelligence in Health

ORIGINAL RESEARCH ARTICLE
Leveraging summary of radiology reports with

transformers

Raul Salles de Padua * and Imran Qureshi *
1
2
1 Quod Analytics, Niterói, Rio de Janeiro, Brazil
2 Department of Computer Science, University of Texas Austin, Austin, Texas, United States of
America

Abstract

Two fundamental problems in health-care stem from patient handoff and triage.
Doctors are often required to perform complex findings summarization to facilitate
efficient communication with specialists and decision-making on the urgency of
each case. To address these challenges, we present a state-of-the-art radiology report
summarization model utilizing adjusted bidirectional encoder representation from
transformers BERT-to-BERT encoder–decoder architecture. Our approach includes
a novel method for augmenting medical data and a comprehensive performance
analysis. Our best-performing model achieved a recall-oriented understudy for
gisting evaluation-L F1 score of 58.75/100, outperforming specialized checkpoints
with more sophisticated attention mechanisms. We also provide a data processing
pipeline for future models developed on the MIMIC-chest X-ray dataset. The model
introduced in this paper demonstrates significantly improved capacity in radiology
*Corresponding authors: report summarization, highlighting the potential for ensuring better clinical
Raul Salles de Padua workflows and enhanced patient care.
(raul.padua@iese.net)
Imran Qureshi
(imranq@utexas.edu) Keywords: Text summarization; Natural language processing; Deep learning; Artificial
Citation: de Padua RS, Qureshi I. intelligence; Health care; Bidirectional encoder representations from transformers;
Leveraging summary of radiology MIMIC-chest X-ray
reports with transformers. Artif Intell
Health. 2024;1(4):85-96.
doi: 10.36922/aih.3846
Received: June 4, 2024 1. Introduction
Accepted: August 5, 2024 Text summarization helps people devote attention to the most important parts of books,
Published Online: September 26, large bodies of text, and documents. In the process of radiology reporting, doctors
2024 need to painstakingly summarize complex findings to facilitate communication with
Copyright: © 2024 Author(s). specialists, and the resulting technical reports, which are typically long, are largely
This is an Open-Access article obscure for most patients. The task of report summarization is an extremely crucial part
distributed under the terms of the
Creative Commons Attribution of radiology reporting, but the complexities and challenges involved in reporting are
License, permitting distribution, further compounded by a shortage of practicing radiologists in the United States. 1
and reproduction in any medium,
provided the original work is Despite the advances attained, natural language processing (NLP) has rarely been
properly cited. applied to medical summarization tasks, particularly in radiology; therefore, very little is
Publisher’s Note: AccScience known about language models that are specifically trained for radiology summarization.
Publishing remains neutral with Radiology reports are typically lengthy and filled with technical jargons, making them
regard to jurisdictional claims in
published maps and institutional difficult for non-specialists to interpret. Patients, in particular, may find it challenging
affiliations. to understand their medical conditions and treatment plans after reading these detailed

Volume 1 Issue 4 (2024) 85 doi: 10.36922/aih.3846