Abstract

Abstract In the management of brain tumors, comprehensive genomic profiling testing is used to identify genetic mutations and understand their clinical significance to facilitate optimal therapeutic strategies. However, the test can identify non-standard mutations for brain tumor classification, and accurately presenting their clinical relevance is challenging. Providing accurate clinical trials per patient requests based on these results can also be time-consuming, creating the need for a physician support system. Large Language Models (LLMs) such as GPT, which are used for general conversation, question answering, and text generation, have limitations when used in specialized domains such as medicine. The information derived only from training data results in the inability of the LLM to handle specialized knowledge queries and verify certain facts, increasing the risk of generating incorrect information. To overcome these challenges, we developed a system that integrates a retriever with LLM. The retriever acts as an external memory, allowing LLM to access information beyond the training data. Because the retriever uses human-verified information, it mitigates the potential for incorrect information generation by LLM.In this presentation, we will discuss the use of this system in actual brain tumor diagnosis and treatment, and evaluate its accuracy and reliability. We believe that this system can help diagnose and treat patients with various genetic mutations, not limited to brain tumors.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call