2024-05-21
Large Language Models in Precision Oncology
- What they can and cannot do
- What they can and cannot do
Dr. Manuala Benary
📍 Berlin
👨🏽🔬 Wissenschaftliche Mitarbeiterin
🏫 Core Unit Bioinformatics / Comprehensive Cancer Center Berlin Charié / BIH
Inhalte
Large Language Models (LLMs) are AI tools trained to generate new text from large datasets. Since the release of ChatGPT, they have received significant public attention and sparked a debate about the benefits and risks of Artificial Intelligence to an unprecedented degree. LLMs enable tasks such as answering questions, summarizing text, or translating, with a quality that approaches the human level. The use of LLMs is increasingly entering the medical field, changing how research is conducted on medical issues, how medicine is taught, and how it is practiced by physicians.
In this presentation, we will give an introduction into LLMs and present a study, which evaluates the potential of conversational large language models (LLMs) as tools for personalized decision-making in precision oncology. Ten fictional cases of advanced cancer with genetic alterations were assessed by four different LLMs (ChatGPT, Galactica, Perplexity, and BioMedLM) and one expert physician to identify treatment options. While LLMs provided a higher number of treatment options compared to the expert, their suggestions often deviated from expert recommendations. However, LLMs did propose some reasonable strategies not easily identified by experts, suggesting potential for improvement in LLM-based decision-making tools. Despite not yet being suitable for routine clinical use, LLMs may enhance existing methods in oncology decision support, particularly as technology advances.
Lernziele
At the conclusion of this session, participants should be able to:
1. Name different Large Language Models which are currently available.
2. Know the potential applications of LLMs in health care.
3. Understand possible constraints of LLMs in the field of precision medicine.
Folien
202405_LLMsInMolecularTumorBoards.pdf