By Dr. Marco V. Benavides Sánchez.
Artificial intelligence (AI) has ceased to be a futuristic promise and has become an everyday tool across multiple fields. Among its most notable advances are large language models (LLMs), capable of understanding and generating text with remarkable fluency. While many associate them with virtual assistants or content generators, their impact on biomedicine is growing rapidly, opening new possibilities for research, diagnosis, and clinical practice.
A recent study published in Artificial Intelligence in Medicine offers a comprehensive overview of this phenomenon. Unlike previous reviews focused on specific applications, this work analyzes more than 480 scientific publications to map the current state, challenges, and future prospects of LLMs in the biomedical domain.
What are LLMs and why are they relevant to medicine?
LLMs are AI systems trained on massive volumes of text. They don’t merely repeat information—they can adapt to different contexts, answer complex questions, generate hypotheses, and make inferences. In medicine, where knowledge is conveyed through scientific articles, clinical records, protocols, and genomic databases, this capability is especially valuable.
Imagine a model that can read thousands of studies, cross-reference clinical data, and suggest potential diagnoses or personalized treatments. That’s the promise now beginning to take shape.
Current Applications: Beyond the Laboratory
The study identifies several areas where LLMs are already delivering tangible results:
- Diagnostic assistance: LLMs analyze symptoms, clinical notes, and lab results to offer differential diagnoses. They don’t replace physicians, but they can accelerate decision-making and reduce errors.
- Drug discovery: By reviewing scientific literature and molecular databases, these models detect patterns that suggest new therapeutic combinations or alternative uses for existing medications.
- Personalized medicine: By cross-referencing genetic, clinical, and pharmacological data, LLMs help predict which treatment is most likely to work for each individual patient.
- Biomedical literature processing: With millions of articles published each year, these models can summarize, classify, and extract key information, making it easier for researchers to stay up to date.
Learning Without Specific Training
One notable feature is the so-called zero-shot learning, which allows LLMs to perform tasks without having been specifically trained for them. For instance, a general-purpose model can answer medical questions without having been fed specialized literature.
This greatly expands their applicability, but it also presents limitations. In areas that demand high precision—such as interpreting medical images or responding to complex clinical questions—fine-tuning is necessary to enhance their performance.
Smart Adaptation: Text, Images, and More
To optimize LLMs in biomedical contexts, the study proposes two key strategies:
- Specialized fine-tuning: involves retraining the model with specific medical data, such as electronic health records or literature from a particular specialty.
- Multimodal models: integrate text, images, genetic sequences, and numerical data, enabling richer understanding and more comprehensive responses.
Ethical and Technical Challenges: How to Move Forward Responsibly
Like any groundbreaking and paradigm-shifting technology, LLMs face significant challenges before they can be widely adopted in hospitals and laboratories:
- Data privacy: Medical information is highly sensitive. How can models be trained without compromising patient confidentiality?
- Limited interpretability: Many models operate as “black boxes,” providing answers without clearly explaining how they arrived at them. In medicine, this lack of transparency breeds distrust.
- Data quality: If training datasets contain biases or errors, models may replicate and even amplify them.
- Ethical responsibility: Who is accountable if a model suggests an incorrect diagnosis? The legal implications are still being defined.
Conclusion: A Revolution Underway
Language models in biomedicine are still in the development phase, but their evolution is swift. What today seems like a promise could soon become part of everyday medical practice. From accelerating research to personalizing treatments, LLMs have the potential to transform modern medicine.
The challenge will be to harness their power without losing sight of patient safety, ethics, and dignity. If achieved, we could be witnessing now a silent revolution that will shape the course of 21st-century medicine.
Further Reading:
- Wang, C., Li, M., He, J., Wang, Z., Darzi, E., Chen, Z., Ye, J., Li, T., Su, Y., Ke, J., Qu, K., Li, S., Yu, Y., Liò, P., Wang, T., Wang, Y. G., & Shen, Y. (2025). A survey for large language models in biomedicine. Artificial Intelligence in Medicine, 103268. https://doi.org/10.1016/j.artmed.2025.103268
#ArtificialIntelligence #Medicine #Surgery #Medmultilingua
Leave a Reply