Leveraging Summary Guidance on Medical Report Summarization
This work addresses automated summarization for medical reports, which is incremental as it builds on existing pre-trained models with guidance techniques.
The study tackled medical report summarization by introducing three large datasets and a method that uses sampled summaries as guidance to enhance a BART model, resulting in improved ROUGE and BERTScore metrics that outperformed a larger T5-large model.
This study presents three deidentified large medical text datasets, named DISCHARGE, ECHO and RADIOLOGY, which contain 50K, 16K and 378K pairs of report and summary that are derived from MIMIC-III, respectively. We implement convincing baselines of automated abstractive summarization on the proposed datasets with pre-trained encoder-decoder language models, including BERT2BERT, T5-large and BART. Further, based on the BART model, we leverage the sampled summaries from the train set as prior knowledge guidance, for encoding additional contextual representations of the guidance with the encoder and enhancing the decoding representations in the decoder. The experimental results confirm the improvement of ROUGE scores and BERTScore made by the proposed method, outperforming the larger model T5-large.