Imagine a world where artificial intelligence can seamlessly interpret medical scans in any language, even with limited data. This is the challenge tackled by researchers in a new study focused on classifying radiological text in Danish, a low-resource language. Why is this so important? Radiology reports are a goldmine of clinical information, but they're locked away in unstructured text. Manually labeling these reports is time-consuming and expensive, especially when dealing with rare conditions and imbalanced datasets. This is where AI steps in. The research explored various NLP models, including BERT-like transformers, few-shot learning with sentence transformers (SetFit), and large language models (LLMs). Surprisingly, the simpler BERT-like models, especially those pre-trained on a large dataset of Danish radiology reports, outperformed the more complex LLMs and SetFit models. While none of the models achieved perfect accuracy, they showed promise in pre-screening reports, potentially reducing the workload on medical professionals. This research highlights a key challenge in applying AI to global healthcare: the need for robust models that can handle diverse languages and limited data. The findings pave the way for more efficient use of medical data, ultimately improving diagnosis and treatment for patients worldwide.