05 Dec 2024
The CLINICSUM framework employs a two-module architecture, combining retrieval-based filtering with fine-tuned open-source large language models, to automatically generate accurate SOAP-format clinical summaries from patient-doctor conversations. When utilizing LLAMA-3-8B, the framework achieves a ROUGE-1 F1-score of 0.70, outperforming proprietary GPT-4-Turbo (0.58), and is preferred by medical experts in 61% of cases over GPT-generated summaries.
04 Mar 2025
The healthcare industry is moving towards a patient-centric paradigm that requires advanced methods for managing and representing patient data. This paper presents a Patient Journey Ontology (PJO), a framework that aims to capture the entirety of a patient's healthcare encounters. Utilizing ontologies, the PJO integrates different patient data sources like medical histories, diagnoses, treatment pathways, and outcomes; it enables semantic interoperability and enhances clinical reasoning. By capturing temporal, sequential, and causal relationships between medical encounters, the PJO supports predictive analytics, enabling earlier interventions and optimized treatment plans. The ontology's structure, including its main classes, subclasses, properties, and relationships, as detailed in the paper, demonstrates its ability to provide a holistic view of patient care. Quantitative and qualitative evaluations by Subject Matter Experts (SMEs) demonstrate strong capabilities in patient history retrieval, symptom tracking, and provider interaction representation, while identifying opportunities for enhanced diagnosis-symptom linking. These evaluations reveal the PJO's reliability and practical applicability, demonstrating its potential to enhance patient outcomes and healthcare efficiency. This work contributes to the ongoing efforts of knowledge representation in healthcare, offering a reliable tool for personalized medicine, patient journey analysis and advancing the capabilities of Generative AI in healthcare applications.
18 Mar 2025
Research from Mississippi State University, the University of Alabama, and Potentia Analytics Inc. developed an automated framework to construct Patient Journey Knowledge Graphs (PJKGs) from structured patient data and unstructured clinical conversations using Large Language Models (LLMs). The approach achieved 100% structural compliance with a predefined ontology, with Anthropic's Claude 3.5 demonstrating the highest semantic accuracy (F1-score of 0.73) for entity and relationship extraction, while Mistral exhibited superior computational efficiency and scalability.
There are no more papers matching your filters at the moment.
