A GEN AI Framework for Medical Note Generation

📅 2024-09-27
🏛️ 2024 6th International Conference on Artificial Intelligence and Computer Applications (ICAICA)
📈 Citations: 6
Influential: 0
📄 PDF
🤖 AI Summary
Physician burnout and reduced clinical time are exacerbated by excessive documentation burdens in healthcare. To address this, we propose MediNotes—a novel framework that, for the first time, integrates lightweight automatic speech recognition (ASR), retrieval-augmented generation (RAG), and quantized low-rank adaptation (QLoRA)-based parameter-efficient fine-tuning (PEFT) to enable real-time, structured SOAP note generation from physician–patient dialogues (speech or text). The architecture supports query-driven clinical information retrieval while ensuring high accuracy, low latency, and compatibility with edge or offline deployment. Evaluated on the ACI-BENCH benchmark, MediNotes significantly improves note accuracy, generation efficiency, and clinical utility—reducing documentation time by over 60%. This work provides a practical, deployable solution for alleviating electronic health record (EHR) documentation burdens and advancing intelligent primary care.

Technology Category

Application Category

📝 Abstract
The growing administrative demands of medical documentation, particularly through Electronic Health Records (EHR), have substantially reduced the time available for direct patient care and exacerbated physician burnout. To mitigate this challenge, we introduce MediNotes, an advanced generative AI framework designed to automate the creation of SOAP (Subjective, Objective, Assessment, Plan) notes from medical conversations. MediNotes leverages Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and Automatic Speech Recognition (ASR) to process both textual and voice inputs in real time or from recorded audio, generating structured, contextually accurate medical notes. The framework employs cutting-edge techniques such as Quantized Low-Rank Adaptation (QLoRA) and Parameter-Efficient Fine-Tuning (PEFT) to optimize model performance in resource-limited settings. Furthermore, MediNotes features a query-based retrieval system, enabling healthcare providers and patients to efficiently access relevant medical information. Evaluation on the ACI-BENCH dataset demonstrates that MediNotes enhances the accuracy, efficiency, and usability of automated medical documentation, presenting a robust solution to alleviate administrative burdens on healthcare professionals while improving the quality of clinical workflows.
Problem

Research questions and friction points this paper is trying to address.

Automating SOAP note generation from medical conversations
Reducing administrative burden and physician burnout from EHR documentation
Integrating LLMs, RAG and ASR for accurate medical documentation
Innovation

Methods, ideas, or system contributions that make the work stand out.

Uses LLMs with RAG and ASR integration
Employs QLoRA and PEFT for efficient fine-tuning
Provides query-based retrieval for medical information
🔎 Similar Papers
No similar papers found.
H
Hui Yi Leong
Data Science Institute, University of Chicago, Chicago, Illinois, USA
Y
Yi Fan Gao
Data Science Institute, University of Chicago, Chicago, Illinois, USA
S
Shuai Ji
Data Science Institute, University of Chicago, Chicago, Illinois, USA
B
B. Kalaycıoğlu
University of Chicago Medical, University of Chicago, Chicago, Illinois, USA
U
Uktu Pamuksuz
Data Science Institute, University of Chicago, Chicago, Illinois, USA