Zero-shot Thoracic Oncologic History Generation for Radiologists Using Retrieval-augmented Large Language Model Pipeline

The authors aimed to develop and evaluate a zero-shot large language model (LLM) pipeline for generating structured oncologic histories to improve efficiency in gathering clinical summaries for thoracic oncology patients. Their study found that the GPT-5-mini model achieved the highest completeness and accuracy in summarization, while also demonstrating significant time savings compared to traditional manual methods, potentially leading to increased revenue for radiologists. Overall, the LLM pipeline effectively summarizes oncologic history without the need for manual fine-tuning or information retrieval.