From Data to Insights: A Survey on Biomedical Text Summarization Approaches and Challenges

Abstract

The explosive growth of biomedical literature and clinical data is increasing the difficulty for healthcare professionals to continuously access new information. This survey summarizes the state-of-the-art studies in biomedical text summarization (extractive, abstractive, and hybrid) and their implementations. In this regard, the survey delves into the influence of deep learning and natural language processing (NLP) methods on enhancing summarization capabilities while simultaneously highlighting ongoing challenges related to domain-specific jargon, truthfulness, and explainability. The paper further describes some evaluation metrics and datasets specific to biomedical problems that are important for both training and evaluating summarization models. Finally, we highlight three of these considerations that we believe are worth pursuing in future work toward designing more effective and grounded summarization systems. Biomedical text summarization, which extracts useful information from vast quantities of data, might help render scientific knowledge more reachable, assist clinical decision-making and policy, and potentially push research forward.