Deep Learning-Based Brain Hemorrhage Detection in CT Reports.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Radiology
• Natural Language Processing in Healthcare

Background:

• Radiology reports contain critical information for patient care.
• Automated detection of critical findings, such as brain hemorrhage, is valuable.
• Computed Tomography (CT) scans are frequently used for brain imaging.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting brain hemorrhage from Turkish radiology reports.
• To assess the impact of pre-trained word representations and domain-specific fine-tuning on model performance.
• To establish the feasibility of using AI for automated critical finding detection in radiology.

Main Methods:

• A deep learning classifier was trained on a large dataset of Turkish radiology reports.
• Various pre-trained word representations were explored.
• Domain-specific fine-tuning of language models was performed using radiology data.
• The model's performance in detecting brain hemorrhage was evaluated.

Main Results:

• The study reports the performance of a large-scale classification model for brain hemorrhage detection.
• Fine-tuning pre-trained language models with domain-specific data significantly improved classification accuracy.
• The developed deep learning model achieved reasonable accuracy in identifying brain hemorrhage.

Conclusions:

• Deep learning models are effective for detecting brain hemorrhage from radiology reports.
• Domain-specific fine-tuning enhances the performance of language models for medical text analysis.
• Automated detection systems can aid physicians in identifying critical cases promptly.