Automated Data Extraction Using Large Language Models in Vascular and Interventional Neurology Research: Towards an AI Fully Assistant Automated Research
Ahmed Shaheen1, Belal Hamed2, Nour Shaheen1, Fabio Dennstädt3, Abdullrahman Aldohni4, Ram Saha5
1Alexandria Faculty of Medicine, 2Faculty of Medicine Al-azahr Cairo university, 3Department of Radiation Oncology, Inselspital, Bern University Hospital and University of Bern, Bern, Switzerland, 4Faculty of medicine, Ain shams University, Cairo Egypt, 5Virginia Commonwealth University
Objective:
To investigate the application of large language models (LLMs) for automated data extraction from research papers, with the goal of developing a fully automated framework for meta-analysis and literature review.
Background:
The integration of Artificial Intelligence (AI) into medical research is revolutionizing healthcare, offering a transformative approach to education and enabling future researchers to push the boundaries of their fields. Specifically, in vascular and interventional neurology research, AI tools hold immense potential for accelerating data analysis, extraction, literature review and other routine teaks. As illustrated in recent literature, AI systems demonstrate the capability to learn complex strategies and sift through vast quantities of data with minimal human intervention. This capability is crucial for developing systems that can autonomously extract key information from unstructured text, such as research articles, ultimately facilitating the creation of a fully automated research framework.
Design/Methods:
we conducted a literature review for “late vs early thrombectomy in large vessel stroke patients”. Literature was then review by two independent researchers. 19 Studies were identified and PDF files were uploaded into a drive. Then PDF files were converted into text and tokenized. Using Gemini Pro Model in a colab notebook we managed extracted all the meta-analysis variables in structured format. .
Results:
The model achieved outstanding results in all tasks with 100% accuracy for all the extracted variables (Study design, sample size, patient demographics, medical history, treatment, early and late neurological outcomes)
Conclusions:
LLM models have reached the maturity and now can be implemented in fully automated clinical and neurological research frameworks. They can achieve outstanding accuracy with propriate supervision. They can reduce need for extra hands and costs in labs and save more time for education, teaching and other tasks
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.