Extraction of immunotherapy clinical features in Non-Small Cell Lung Cancer using privacy-preserving Large Language Models

Aims:

• assess the capabilities of privacy-preserving LLMs, such as Llama and GPT-4 in extracting clinical information from the textual medical data and for specific clinical variables a comparison of LLMs performance with NLP algorithms;
• investigate strategies for leveraging usable Explainable AI (XAI) techniques in the context of LLMs to boost model trustworthiness, including security, fairness and truthfulness, which is crucial to achieve human alignment.

Collaborators: Vittoria Peppoloni, Giuseppe Leone, Justin Armanini, Ilaria Cavallo, Paolo Baili