Fast Rule-based NER in SpaCy for Chest Radiography Reports with CheXpert's 14 Categories

Chest X-ray imaging is widely used in medical examinations, with 2 billion performed globally annually. Interpreting X-ray images is time-consuming, prompting the development of AI-assisted systems. Generating a large set of high-quality labeled images is crucial for training accurate and unbiased deep learning models. However, creating high-quality labeled images is challenging. Utilizing Natural Language Processing (NLP) and Named Entity Recognition (NER), we can expedite this process by extracting diagnostic information from text reports. In this study, we developed a rule-based NER tool in SpaCy that can recognize the words and expressions defined in CheXpert's 14 categories. We tested our software using the U.S. National Library of Medicine Open-i dataset. It achieved equivalent accuracy in labeling medical terms with one-sixth the computational time compared to the CheXpert Labeler. This approach holds promise to improve the efficiency of labeling of X-ray images for training AI models to be more accurate in detecting abnormality.Clinical Relevance- This study applied SpaCy NER to label chest X-ray reports with the 14 CheXpert's diagnostic categories with high accuracy. This could reduce human labeling time and error in generating high-quality training datasets for training automated medical imaging analysis software to reduce the burden on doctors.