This presentation introduces a novel framework for extracting meaningful clinical information from hospital discharge summaries by combining Spark NLP with modern AI models—large language models (LLMs) such as LLaMA3 and Mistral. Unstructured clinical text contains valuable insights that are often difficult to utilize effectively. The proposed approach addresses this challenge by identifying key clinical concepts and linking them to standardized medical terminologies, including UMLS and SNOMED CT.

Spark NLP is used to detect relevant text spans, while LLaMA3 and Mistral enhance contextual understanding and extraction accuracy. Clinical experts evaluated the extracted concepts, and the system showed strong performance—surpassing traditional rule-based tools in handling abbreviations, multi-word expressions, and diverse language patterns.

This talk will demonstrate how the integration of Spark NLP and LLMs enables more accurate, scalable, and context-aware clinical concept extraction. The framework offers a flexible solution to improve data quality, support clinical decision-making, and power downstream healthcare applications.