LLMs Have Promising Potential in Radiology-Related Report Classification

Background: Stroke is the 5th leading cause of death in the United States, the second leading cause of death worldwide, and is the leading cause of acquired long-term disability. Recognition of acute and subacute infarcts is important as prompt therapy may limit morbidity and disability. Imaging modalities play a critical role in the management of stroke, with both MRI and CT used in clinical practice, but in some cases, these radiological examinations and reports are not adequately reviewed, which can delay treatment. Large language models (LLMs) based on Bidirectional Encoder Representations from Transformers (BERT) fine-tuned on radiology-specific datasets, have advanced clinical natural language processing and have promising potential in radiology-related report classification. It is hoped that these models can serve as a warning mechanism, preventing clinicians from overlooking unexpected strokes and facilitate rapid assessment. Objective: To develop an automated early warning s more...