Thoracic Aorta Measurement Extraction from Computed Tomography Radiology Reports Using Instruction Tuned Large...

Area of Science:

• Radiology and Medical Informatics
• Artificial Intelligence in Healthcare
• Cardiovascular Disease Research

Background:

• Chest computed tomography (CT) is crucial for managing thoracic aortic aneurysms and dilations.
• Aortic measurements in radiology reports are often unstructured, hindering clinical use and research.
• Complications like aortic dissection and rupture pose significant risks to patients.

Purpose of the Study:

• To develop a pipeline for extracting structured aortic measurements from free-text chest CT radiology reports.
• To compare the efficacy of BERT-based models against instruction-tuned Llama large language models (LLMs) for this task.
• To create a comprehensive dataset for large-scale aortic disease research.

Main Methods:

• Development of a multi-method pipeline for automated data extraction.
• Comparative analysis of fine-tuned BERT models and instruction-tuned Llama LLMs.
• Application of the optimal extraction method to a large database of chest CT reports.

Main Results:

• Successful extraction of structured aortic measurements from unstructured radiology reports.
• Demonstration of the effectiveness of AI models, particularly LLMs, in processing clinical text.
• Generation of a large-scale, structured dataset of aortic measurements from real-world data.

Conclusions:

• The developed pipeline effectively extracts critical aortic measurements, enhancing data accessibility.
• This structured dataset significantly supports big data research in aortic diseases.
• AI-driven natural language processing offers a powerful solution for unlocking valuable information in clinical reports.