Word Graph Guided Summarization for Radiology Findings

Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4980–4990 August 1–6, 2021. ©2021 Association for Computational Linguistics 4980 Word Graph Guided Summarization for Radiology Findings Jinpeng Hu♠♥, Jianling Li♦, Zhihong Chen♠♥, Yaling Shen♠, Yan Song♠♥, Xiang Wan♥, Tsung-Hui Chang♠♥ ♠The Chinese University of Hong Kong (Shenzhen), ♥Shenzhen Research Institute of Big Data, ♦National University of Defense Technology ♠{jinpenghu, zhihongchen, yalingshen}@link.cuhk.edu.cn ♦jianlingl@nudt.edu.cn ♥wanxiang@sribd.cn ♠{songyan,changtsunghui}@cuhk.edu.cn Abstract Radiology reports play a critical role in com- municating medical findings to physicians. In each report, the impression section summa- rizes essential radiology findings. In clini- cal practice, writing impression is highly de- manded yet time-consuming and prone to er- rors for radiologists. Therefore, automatic impression generation has emerged as an at- tractive research direction to facilitate such clinical practice. Existing studies mainly fo- cused on introducing salient word information to the general text summarization framework to guide the selection of the key content in radiology findings. However, for this task, a model needs not only capture the impor- tant words in findings but also accurately de- scribe their relations so as to generate high- quality impressions. In this paper, we pro- pose a novel method for automatic impres- sion generation, where a word graph is con- structed from the findings to record the critical words and their relations, then a Word Graph guided Summarization model (WGSUM) is designed to generate impressions with the help of the word graph. Experimental results on two datasets, OPENI and MIMIC-CXR, con- firm the validity and effectiveness of our pro- posed approach, where the state-of-the-art re- sults are achieved on both datasets. Further experiments are also conducted to analyze the impact of different graph designs to the perfor- mance of our method.1 1 Introduction A radiology report usually contains a findings sec- tion (FINDINGS) describing detailed medical obser- vations and an impression section (IMPRESSION) summarizing the most critical observations. In practice, the IMPRESSION is an essential part and 1Our code and the best performing models are released at https://github.com/cuhksz-nlp/WGSum. There is a moderate le� pleural effusion, similar to mul�ple recent studies. Le� basilar opacification is similar to the prior exam, likely atelectasis, but infec�on can not be excluded. There is mild pulmonary vascular conges�on and cardiomegaly. There is no pneumothorax. Mild pulmonary vascular conges�on, cardiomegaly and a moderate le� pleural effusion. Persistent le� basilar opa cifica�on may reflect compressive atelectasis but infec�o -n is not excluded. atelectasis similar excluded be not ... ... Findings Impression Radiology Report Graph pleural le� moderate likely infec�on can basilar opacification effusion Figure 1: An example radiology report including its FINDINGS and IMPRESSION, and a word graph con- structed from the FINDINGS. Different colored edges in the graph represent different types of relations. The curved arrow indicates the AIG task to generate the IM- PRESSION from the FINDINGS. plays an important role in delivering critical find- ings to clinicians. Therefore, summarizing FIND- INGS helps to locate the most prominent observa- tions so that the automatic process of doing so greatly eases the workload …

Content truncated to 500 words