Guided Notes

Clinical Natural Language Processing

Why clinical text is hard, NLP tasks (entity recognition, negation, normalization), rule-based vs machine learning vs large language models, evaluation and real-world uses.

Why clinical text is hard

Slide 1 / 5

• Abbreviations and acronyms with multiple meanings ('MS' = multiple sclerosis, mitral stenosis, morphine sulfate). • Negation ('no chest pain'), uncertainty ('possible pneumonia'), history ('h/o MI'), family history ('mother had breast cancer'). • Misspellings, telegraphic style, copy-paste. • Templates and lists mixed with narrative. A keyword search for 'pneumonia' would wrongly include 'no evidence of pneumonia'.