A Cross-document Coreference Dataset for Longitudinal Tracking across Radiology Reports

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3686–3695 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3686 A Cross-document Coreference Dataset for Longitudinal Tracking across Radiology Reports Surabhi Datta, Hio Cheng Lam, Atieh Pajouhi, Sunitha Mogalla, Kirk Roberts School of Biomedical Informatics, The University of Texas Health Science Center at Houston {surabhi.datta,kirk.roberts}@uth.tmc.edu Abstract This paper proposes a new cross-document coreference resolution (CDCR) dataset for identifying co-referring radiological findings and medical devices across a patient’s radiology reports. Our annotated corpus contains 5872 mentions (findings and devices) spanning 638 MIMIC-III radiology reports across 60 patients, covering multiple imaging modalities and anatomies. There are a total of 2292 mention chains. We describe the annotation process in detail, highlighting the complexities involved in creating a sizable and realistic dataset for radiology CDCR. We apply two baseline methods–string matching and transformer language models (BERT)–to identify cross-report coreferences. Our results indicate the requirement of further model development targeting better understanding of domain language and context to address this challenging and unexplored task. This dataset can serve as a resource to develop more advanced natural language processing CDCR methods in the future. This is one of the first attempts focusing on CDCR in the clinical domain and holds potential in benefiting physicians and clinical research through long-term tracking of radiology findings. Keywords: tracking, cross-document coreference resolution, radiology 1. Introduction Radiology reports contain rich descriptions of clini- cally important findings and medical devices. Often- times, these findings and devices are referred to mul- tiple times in a single report and are also referred to across different reports of a patient. Radiologists make such references in multiple reports mainly to highlight any longitudinal changes of a particular finding (e.g., change in a tumor at a certain location) and also to de- scribe any interval changes in a device position (e.g., change in the position of an endotracheal tube inserted in a patient with respect to an anatomical location). Although extracting important information (e.g., find- ings, anatomical locations) from radiology reports has been widely studied (Hassanpour and Langlotz, 2016; Steinkamp et al., 2019; Datta et al., 2020; Syeda- Mahmood et al., 2020; Sugimoto et al., 2021), tracking (or identifying the coreferences) of radiological find- ings across reports is unexplored. Automated track- ing of findings and devices across a patient’s radiology reports holds potential to reduce physician burden in making patient-related decisions as well as to facilitate various retrospective clinical research studies. Tracking the same finding or device across re- ports is a challenging problem as it relies on radi- ology domain knowledge and requires understanding the linguistic variations used by radiologists as well as understanding both linguistic and domain-specific context across different reports. This is illustrated through an example in Figure 1, where, for Patient 1, the perihilar edema described in one of the subse- quent reports of this patient is referencing to the pul- monary edema mentioned in a previous report, and is again described through a different …

Content truncated to 500 words