A New Dataset for Summarizing Radiology Reports

Anonymous

Abstract: The radiology report summarization is an important technology in smart healthcare. Compared with medical image processing and disease recognition which have been comprehensively studied, the research on radiology report summarization is much limited, which is mainly due to the lack of a high-quality benchmark dataset. In this paper, we present a dataset called CRRsum for radiology report summarization, where it is constructed from over 10K real radiology reports that contains diagnostic findings and diagnostic opinions. An extensive evaluation is performed with the current state-of-the-art methods for radiology report summarization on our proposed dataset. Our experiments reveal the challenges of radiology report summarization and provide many opportunities for research going forward. We also show that the CRRsum can be used in medical classification to facilitate the research in this task.

Paper Type: long