Large Language Models for Classifying Usual Interstitial Pneumonia from Radiology Reports: Native Reasoning Versus Structured Prompting.

While prior studies have evaluated large language models for radiology report labeling, none have compared how prompting strategies interact with the native reasoning capabilities of newer model architectures.

We evaluated 10 open-source large language models from three architecture families (Llama, Qwen, Gemma) spanning 8 to 405 billion parameters, each tested with three prompting strategies on 270 high-resolution computed tomography reports classified by expert consensus of two senior thoracic radiologists.